Preprints.org logo
Preprint
Review

This version is not peer-reviewed.

Comparing Experimental and Computational Approaches to Identify RNA-Protein Interactions

Submitted:

16 July 2026

Posted:

17 July 2026

You are already at the latest version

Abstract
RNA-binding proteins (RBPs) are essential regulators of RNA metabolism and gene expression, influencing processes such as splicing, stability, localization, and translation. Despite their critical roles in health and disease, including cancer, identifying RNA-protein interactions remains challenging due to technical limitations and biases of existing methods. Here we review and compare experimental techniques—including in vitro affinity purification, in vivo cross-linking, and proximity labeling—and computational prediction tools for RBP identification. We assess their strengths, limitations, and applicability across biological contexts, emphasizing the benefits of integrating experimental and computational strategies. Our analysis provides practical guidelines for selecting appropriate methodologies tailored to different cell types and research goals. These insights aim to facilitate more accurate mapping of RNA-protein interactomes, thereby advancing understanding of RBP functions and supporting the development of novel therapeutic interventions targeting RNA-protein complexes.
Keywords: 
RNA-binding proteins
;  
regulatory RNAs
;  
noncoding RNAs
;  
RNA-protein interactions
;  
computational prediction methods
;  
experimental identification techniques
;  
RNA stability and localization
Subject: 
Biology and Life Sciences  -   Biochemistry and Molecular Biology

1. Introduction

Within the cell, RNA molecules are almost always bound by numerous RNA-binding proteins (RBPs), which play essential roles in protecting and regulating RNA. RBPs constitute approximately 5-10% of the human proteome, and over 4000 potential RBPs have been identified through computational predictions and mass spectrometry approaches, although a majority of these remain functionally uncharacterized (Biswas et al., 2020; Konde et al., 2025; Sohrabi-Jahromi & Söding, 2021; Yang et al., 2021).
RBPs regulate a wide range of cellular processes, including RNA processing, export, localization, translation, decay, splicing, polyadenylation, stability, and interactions with
ther partners (Biswas et al., 2020; Konde et al., 2025; Liang et al., 2024; Lindsey et al., 2026; Nicholson-Shaw et al., 2022; Shen et al., 2022; Sohrabi-Jahromi & Söding, 2021; Street et al., 2024). They also play roles in the folding and assembly of ribosomal RNA-protein complexes and the processing of small nuclear RNAs (Lindsey et al., 2026). Figure 1 illustrates an example of the mRNA life cycle with a particular emphasis on miRNA biogenesis to highlight the diverse functional roles of RNA binding proteins throughout RNA metabolism.
RBPs interact with both coding and non-coding RNAs, such as microRNAs (miRNAs), small interfering RNAs (siRNAs), small nuclear RNAs (snRNAs), long-noncoding RNAs (lncRNAs), and circular RNAs (circRNAs), to control RNA function (Jeffrey et al., 2025; Konde et al., 2025; Shen et al., 2022; Sommerauer & Kutter, 2022; Tang et al., 2025). Notably, RBPs have been reported to influence circRNA biogenesis and catabolism (Tang et al., 2025). When circRNAs interact with RBPs, they can modulate the function of these proteins by acting as decoys or transporters; in some cases, circRNAs can also “sponge” RBPs, thereby modulating their activity (Fu et al., 2022; L. Liu, Y. Wei, Z. Tan, et al., 2024).
Although much progress has been made in identifying RBPs, the precise mechanism by which they selectively bind their RNA targets remains only partially understood (Konde et al., 2025; Xu & Cui, 2025). Despite the diverse structures and functions of RBPs, they typically contain at least one RNA-binding domain (RBD), a conserved protein domain that enables recognition of specific RNA sequences, motifs, secondary structures, or chemical modifications (Biswas et al., 2020; Hao et al., 2023; Kodavati et al., 2024; Konde et al., 2025; Oberstrass et al., 2024; Sohrabi-Jahromi & Söding, 2021). Many RBPs possess multiple RBDs or form complexes with other RBPs, allowing them to bind RNA in a multivalent and cooperative manner, which confers higher specificity and affinity compared to single RBDs; notably, bivalent binding is a common strategy among RBPs that enhances both affinity and sequence specificity (Sohrabi-Jahromi & Söding, 2021).
Intrinsically disordered regions (IDR), such as arginine-glycine-glycine (RGG) motif and tyrosine-rich regions, are also prevalent in RBPs and contribute to RNA binding. Due to their lack of stable structure, structural studies have been focused primarily on the more ordered parts of RBPs, and many RBPs still lack well-characterized RBDs. In addition to an RBD, the most well-known RBPs also include K homology domains (KH), double-stranded RNA-binding domain (dsRBD), DEAD (Asp-Glu-Ala-Asp) box helicase domains, cold-shock domain (CSD), zinc fingers (ZnF), pumilio homology domain (PHD), and IDRs themselves (Hao et al., 2023; Kodavati et al., 2024; Konde et al., 2025; Lindsey et al., 2026; Sommerauer & Kutter, 2022). Interactions between RBDs and RNA occur through diverse mechanisms, such as hydrogen bonds, van der Waals forces, salt bridges, aromatic stacking, and hydrophobic interactions.(Hao et al., 2023; Xu & Cui, 2025).
RBPs have increasingly been recognized as key players in disease development, including genetic diseases, cancer, inflammatory diseases, metabolism-related diseases, and neurodegenerative disorders, and are promising targets for drug discovery, as shown in Supplementary Table 1(Hao et al., 2023; Konde et al., 2025; Liang et al., 2024; Lindsey et al., 2026; Nabeel-Shah & Greenblatt, 2023; Yin et al., 2023; Yu et al., 2025). Genetic mutations in RBPs and abnormal expression have been detected in multiple human tumor types, where cancer cells exploit the post-transcriptional gene regulation of RBPs to modulate protein expression levels and sustain oncogenic proliferation (Lindsey et al., 2026; Shen et al., 2022; Tankka et al., 2025). Although the roles of splicing-regulatory RBPs in apoptosis, angiogenesis, proliferation, and metastasis are increasingly recognized, thousands of novel RBPs in cancer remain poorly understood. Moreover, RBPs can switch between oncogenic and tumor-suppressor roles depending on the specific context within cancer cells (Xu & Cui, 2025). Therefore, new methodologies that enable characterization of roles of RBPs in cancer and identification of their binding sites are still needed to deepen our understanding of tumor biology and identify new therapeutic targets (Tankka et al., 2025; Yang et al., 2021; Yu et al., 2025). Developing RBP inhibitors remains challenging due to the conformational flexibility of RBPs and the lack of well-defined binding pockets (Xu & Cui, 2025).
In recent years, several databases of putative RBP binding sites and partners have been compiled to facilitate RBP identification and to help narrow down target lists prior to experimental validation. However, these databases include only a small number of experimentally validated RBPs and their respective targets (Lasantha et al., 2024), highlighting the need for continued experimental efforts.
With the current rise in the use of RBPs as therapeutic targets, methods to discover and validate binding partners are crucial for developing novel treatments. One of the most commonly used identification methods is CLIP (Cross-Linking and Immunoprecipitation), which requires UV crosslinking of RNA-protein complexes (Srivastava et al., 2021) for identifying RNA-RBP interactions. It employs UV crosslinking prior to cell lysis to prevent adventitious interactions, such as those between cytoplasmic RBPs and nuclear RNAs in different biological contexts (Biswas et al., 2020; Liang et al., 2024; Street et al., 2024; Yu et al., 2025). Since CLIP relies on antibodies, it has the potential to result in high background due to nonspecific antibody-antigen interactions. The method is labor-intensive, requires significant financial resources and demands substantial starting material because of low crosslinking efficiency and the loss of RNA during immunoprecipitation (Biswas et al., 2020; Hao et al., 2023; Liang et al., 2024; Sommerauer & Kutter, 2022; Tang et al., 2025; Xu & Cui, 2025). Additionally, the RNA digestion step of the protocol complicates the identification of RNA isoforms and makes it challenging to determine the dynamics of ribonucleoprotein complexes (RNP)-complexes (Liang et al., 2024; Street et al., 2024; Yu et al., 2025). Therefore, new methods are needed to identify high-confidence targets without sample amount as a limiting factor and in endogenous biological contexts (Biswas et al., 2020; Liang et al., 2024; Yu et al., 2025).
Despite its widespread use, UV crosslinking has been reported to introduce biases into interaction profiles due to its intrinsic chemical properties (Srivastava et al., 2021). Alternatively, antibody-based pulldowns of protein-RNA complexes, such as RIP-seq, are limited to a narrow set of interactions due to their lack of scalability for high-throughput analysis (Srivastava et al., 2021). Other common methods, including poly-A pulldown, are unable to capture non-polyadenylated species such as lncRNAs, miRNAs, and histone mRNAs (Srivastava et al., 2021). Proximity-based labeling enzyme techniques can also yield false positives by capturing proteins that are merely in the vicinity of the RNA of interest rather than directly interacting with it (Dodel et al., 2024). Both proximity biotinylation and UV crosslinking methods are prone to false positives and false negatives, highlighting the need for more stringent identification approaches (Jansen et al., 2021). Furthermore, in situ observations of RBP interactions remain challenging because many methods require the use of cell lysate (Yang et al., 2024). Many of these approaches are also costly and time-consuming, which hinders the discovery and validation of RNA-RBP interactions (Zhou et al., 2023). Collectively, these limitations underscore the urgent need for new methods to identify RNA-protein binding partners, as they create a bottleneck for advancing novel treatments.

1.1. Necessity of RBP Methods

Although our understanding of RNA-binding protein function has evolved, current methods have not yet fully captured the dynamic nature of RBP regulation in cancer. Techniques such as CLIP-seq and RIP-seq, while widely used, lack a comprehensive approach and are limited to providing static snapshots; methods capable of capturing the dynamic interactions of RBPs within tumors are still needed.
Recent developments have shown that novel sequencing methodologies can be used to directly map in vivo RBP-RNA interactions using crosslinking, purification, and ligation techniques. These interactions are regulated by cooperative and competitive binding events that govern gene expression. These approaches help bridge the gap between biochemical accuracy and physiological relevance (Melamed, 2020).
The use of a broad range of testing methods and advanced tools will help better understand the complexity of RBP interactions, as these approaches can capture changes in binding partners in response to cellular stress, development, or cancer-related signals (Sun et al., 2024).
Emerging modeling techniques and binding predictions utilizing machine learning are now being used to predict RBP preferences and binding rules, like those of Pumilio-family proteins. However, the accuracy of such models depends on the quality and diversity of the available experimental data; hence, continued methodology development is greatly needed (Sadee et al., 2022).
However, methodological advances also open clinical perspectives. For example, the development of peptide nucleic acids that mimic the structure of RNAs and inhibit RBPs presents new tools for probing and modulating RBP-RNA interactions, thereby offering potential therapeutic applications (Zhong et al., 2024).
In the long run, integrating approaches such as high-resolution sequencing, structural modeling, and live-cell imaging may facilitate the transition from static catalogs to dynamic networks of RBP interactions and could provide insight into how RBP complexes are reorganized in real time during tumor development or during resistance to therapy. Further development of these tools is not only a technical advance but may also be important for understanding how RNA-protein interactions support cancer cell identity and for translating this knowledge into clinical benefit.

2. Methods to Identify RBPs

RNA-Binding Protein (RBP) identification may be achieved through different methodological categories: in vitro, in vivo, in situ, and in silico. In vitro, or “test tube” methods, such as RNA pull-down, involve using purified components to study the relationships between proteins and RNAs. These methods are typically able to provide high-resolution and quantitative data. The experiments are performed under highly controlled conditions and can provide clear results, but a drawback is the lack of physiological relevance.
On the other hand, in vivo methods are used to capture interactions as they occur naturally within a living cell, rather than in a test tube. They can provide a transcriptome-wide map of RBP targets. Using in vivo methods offers higher biological relevance, but are often associated with more complex protocols, require specific antibodies, and yield data that may be compromised by background noise.
As illustrated in Figure 2, in vitro techniques rely on RNA-pulldown assays to isolate RBP complexes from cell lysates, whereas in vivo methods are much more diverse, encompassing crosslinking methods to directly capture interacting molecules, proximity-labeling strategies to tag adjacent proteins in their native environment, and enzymatic editing techniques that map interactions by actively modifying nearby RNA bases.
In situ methods represent a “middle ground” between in vivo and in vitro approaches. These techniques study the RNA-RBP complexes through fixed cell or tissue samples. The natural context of the molecules is largely preserved, while the complications and limitations associated with live-cell and in vitro experimentations are minimized.
In silico methods are computational approaches that rely on machine learning and bioinformatics to predict RBP relationships. Most of the data used in these methods are generated via experimental approaches, and the predictions from this data can be confirmed through the same methods. The biggest advantage of these methods is their high throughput and cost-effectiveness. This can allow scientists to predict RBP-RNA interactions by screening entire genomes. On the other hand, a major limitation is that the results of these methods must still be experimentally validated to be considered accurate and reliable.

2.1. In Vitro Methods to Identify RBPs

The identification of RBPs in vitro is often achieved through affinity purification methods, most often through the RNA pull-down assay (Table 1). This method utilizes a specific RNA of interest to "pull down" the interacting proteins from a cell lysate. The main components of this technique include attaching a high-affinity tag, most commonly biotin, to the RNA of interest, followed by purifying the RNA-RBP complexes with Streptavidin-coated beads. This method is effective due to the high affinity between biotin and streptavidin. These steps are then followed by washing, elution, and identification of the proteins, typically via mass spectrometry (MS).
In vitro methods include miTRAP and PRIM-seq. Figure 3 presents the general workflow for PRIM-seq, a method for identifying RBPs and their interacting RNAs de novo by performing two experimental modules: SMART-display and REILIS. The first module, SMART-display, consists of extracting mRNA from input cells, adding a linker to the 3’ of each mRNA, and translating the mRNA to create proteins covalently linked with the mRNA. The second module is REILIS, which stands for reverse transcription, incubation, ligation, and sequencing. The mRNA end of the labeled proteins is converted to cDNA; this complex is then incubated with an RNA library to allow RNA-protein interactions. The RNA and cDNA labels are ligated into a chimeric sequence, forming an RNA-linker-cDNA complex, which is then subjected to paired-end sequencing. The cDNA part reflects the protein, and the RNA reflects the interacting RNA. Finally, PRIM-seq resolves the RNA-protein interactions from the chimeric sequences using bioinformatics. Although PRIM-seq offers a robust system for mapping reconstituted bindings rather than endogenous cellular occupancy, there are some limitations, including random priming, which can lead to truncated proteins, and nucleic acid conjugation, which can affect protein folding and RNA binding. The in vitro conditions cannot fully replicate the complexities of the intracellular milieu, and they may yield false positives and false negatives. Consequently, orthogonal approaches, such as targeted mutagenesis, established RNA-protein interaction assays, and cellular assays that capture native binding, should be used in conjunction with PRIM-seq to validate and confirm the biological significance of the findings (Qi et al., 2025).
The miTRAP (miRNA trapping by RNA in vitro affinity purification) method is another example of an affinity-based approach for RBP identification. This method acts effectively by employing an MS2 tag system. The MS2 tags are attached to the RNA of interest and mixed with amylose resin beads and a fusion protein. The fusion protein consists of the MS2-binding protein and maltose-binding protein. The MS2 binding protein can attach to the RNA of interest and capture its interactions, while the maltose-binding protein has a strong affinity to the amylose resin beads. This interaction immobilizes the RNA, and any unspecific binding can be washed away, leaving the purified RBP-RNA interactions. The RNA-binding proteins can be analyzed through mass spectrometry or western blotting (Jasinski-Bergner et al., 2020).
Table 1. In vitro methods.
Table 1. In vitro methods.
Method Description Advantages Limitations Cost* Time Throughput Citation
miTRAP Uses an MS2 tag system and amylose resin beads for affinity-based purification to identify miRNAs and RBPs interacting with a target RNA. Utilizes commercially available components; employs a simplified cloning strategy. Lacks an endogenous cellular environment, limited primarily to miRNAs, Potential for unspecific binding, $$ 5-7 days Low (Jasinski-Bergner et al., 2020)
PRIM-seq Employs SMART-display to barcode proteins with their corresponding mRNA, alongside REILIS to create dsDNA sequences containing the cDNA of both the mRNA and the interacting RNA. Identifies RBPs and associated RNAs at the genome scale; enables de novo identification of interactions; requires no specific reagents to target specific proteins or RNAs. Lacks ability to replicate native cellular conditions; misses post-translational modifications and protein function; lacks suitability for detecting direct interactions; risks false positives and false negatives. $ 3-5 days High (Qi et al., 2025)
* Cost Scale: Detailed as low ($), moderate ($$), and high ($$$) based on the requirement for specialized reagents, such as individually biotinylated probes versus universal oligos. * Time & Throughput: Estimated turnaround times include cell preparation, cross-linking, purification, and mass spectrometry preparation phases.

2.2. In Vivo Methods to Identify RBPs

Chemical/photoreactive-mediated cross-linking methods have been the gold standard for identification of RBPs in vivo. However, there has been a notable surge in the development of methods involving fusion proteins that have been developed to specifically target a molecule of interest. These fusion proteins involve a proximity-labeling enzyme to attach tags to the interactome, and a protein targeting the molecule of interest.

2.2.1. Cross-Linking-Based Methods to Identify RBPs

Cross-linking methods can use formaldehyde/glutaraldehyde or UV light to covalently bind molecules to each other, like RNA and proteins (Table 2). To enhance the capture of RNA-protein interactions, both cross-linkers require high input cell numbers, from 108 to 109 cells. some RNA-centric MS pulldown methods for low-abundance targets historically required very high inputs, with input usually needed to be scaled according to RNA abundance and readout choice. To only extract the RNA-protein cross-linked complexes, RNA is purified under denaturing conditions to remove noncovalent interactions, and cross-linked proteins are extracted for subsequent identification.(Ramanathan et al., 2019).
UV light, compared to formaldehyde, is a zero-distance cross linker for direct contacts, while aldehydes readily generate protein-protein bridges that could blur direct contacts. Nonetheless, cross-linking with UV light is generally less efficient. Also, UV cross-linking has a bias towards uridine, and double-stranded RNA (dsRNA) is poorly cross-linked. The efficiency of UV cross-linking varies by amino acid and may also depend on the structure and surface area of RNA-protein complexes (Ramanathan et al., 2019).
ormaldehyde is a small aldehyde that acts as a bifunctional cross-linker, permeating cells and covalently linking macromolecules within 2 Å in a reversible manner. Formaldehyde cross-linking has a bias toward nucleophilic lysine residues. This compound also forms covalent bonds between proteins in addition to RNA-protein complexes, making it difficult to distinguish between proteins that interact directly with RNA and those that form complexes with bound proteins (Ramanathan et al., 2019).
Among the methods described above, MORPH-MS (Multiple Oligo-assisted RNA Pulldown via Hybridization followed by Mass Spectrometry) and LEAP-RBP (Liquid Emulsion–Assisted Purification of RNA-Bound Protein) are examples of newer approaches that are cost-effective and simple in vivo-approaches to identify RBPs. MORPH-MS is a cross-linking method used to purify and identify proteins associated with a target RNA within the cell (Figure 4). The method relies on an array of 20-25 antisense oligos that are collectively complementary to the entire length of the target RNA. These tiling oligos are designed to provide one oligo per 350-450 nucleotides of RNA length, maintaining a GC content of approximately 48-52%. Their specificity is manually screened using NCBI blasts to select probes with an E-value below 0.0002. Additionally, all oligos share a common sequence at their 5’ end. This common binding sequence allows all antisense oligos to hybridize to a single biotinylated universal oligo. Experimental controls for this method include dividing the probes into separate odd and even groups, incorporating a RNase A control, and utilizing non-target oligos against LacZ. During the procedure, these probes are mixed with the cell lysate, where they bind to the target RNA, and capture the native protein interactions. The biotinylated universal oligo facilitates the purification of these complexes using streptavidin-coated magnetic beads followed by stringent washing steps. Finally, the elution step is performed with benzonase, and the proteins are identified by mass spectrometry. The use of only one biotinylated probe allows for a significant reduction in the price of the procedure (Pant & Kumarswamy, 2024).
The LEAP-RBP method offers another cross-linking based approach to capture the proteins that interact with RNA inside of living cells. Following the cross-linking, the cells are lysed in a guanidinium-based buffer. The lysate is then used to perform an acidic phenol–chloroform extraction, separating the material into distinct layers. The interphase layer, where the RNA-protein complexes are found, is collected and enriched by a lithium-based precipitation step. The purpose of this step is to selectively pull down the crosslinked RNA-protein products and eliminate any free proteins and other potential contaminants. The product is then treated with DNase to remove DNA, leaving only RNA-bound products. To complete the protocol, the proteins are analyzed by mass-spectrometry to be able to be identified. This method is suitable for mapping the entire RNA-protein proteome, but a key limitation is its inability to focus on a single target RNA (Kristofich & Nicchitta, 2024).

2.2.2. In Vivo Proximity-Labeling Methods to Identify RBPs

Proximity labeling (PL) techniques were introduced for the first time in 2012 as an alternative in vivo method to map molecular interactomes (Figure 5 and Table 3). Since then, various PL enzymes and methods have been developed to elucidate protein-protein, DNA-protein, and RNA-protein interactions. The principle behind this set of techniques is similar. A fusion protein that includes the PL enzyme and a protein that can be guided to the RNA of interest is generated. Once the fusion protein has encountered the target RNA, the cell is supplied with a substrate specific to the PL enzyme, which is then covalently attached to the nearby interactome. These molecules are then isolated by affinity-based purification and analyzed by downstream methods depending on what is being studied (Weissinger et al., 2021).
There are two general categories of PL enzymes: peroxidases and biotin ligases. In the presence of hydrogen peroxide (H2O2), peroxidases convert a great variety of substrates into radicals. A widely used peroxidase is APEX2, a plant ascorbate peroxidase variant of APEX engineered to exhibit improved thermal stability and enhanced biotinylating activity (Guo et al., 2023). APEX2 forms free radicals of biotin-tyramide (biotin-phenol; BP) in the presence of biotin-phenol and H2O2. These free radicals react with tyrosine residues on the surface of proteins within a labeling radius of 20 nm, forming covalent adducts. The biotinylated molecules can then be isolated using streptavidin beads and analyzed by downstream methods like mass spectrometry (Kalocsay, 2019).
On the other hand, biotin ligases allow the biotinylating of surrounding molecules. These molecules can then be isolated through biotin affinity purification and identified through various downstream methods; for proteins, this would involve mass spectrometry. There is a wide array of biotin ligases utilized for RBP elucidation, such as BirA*, BASU, BioID2, among others. Each enzyme has their own specifications and labeling radius. BASU, for example, is a 29 kDa enzyme designed from Bacillus subtilis with a 10 nm labeling radius. It allows for the study of RNA-protein interactions in vivo within 1 minute. PL time is crucial to capture transiently interacting proteins, which is why enzymes like BioID2 (16-18 hours) can lead to false-positive or false-negative results. As described before, the biotinylated interactome can then be isolated by and subsequently analyzed based on the needs of the study (Guo et al., 2023).
A PL enzyme that does not fit into these categories is PafA, utilized in CRUIS (presented in Table 3), which attaches PupE to the lysines of surrounding proteins. This pupylation-based interaction tagging (PUP-IT) enriches transient and weakly interacting proteins, detected by mass spectrometry. While the PafA ligase ensures high specificity, Pup is relatively big, which could prevent transmembrane diffusion thereby not allowing for the detection of interactions within organelles (Guo et al., 2023)
Some ways in which the PL enzymes can be targeted to an RNA of interest is through complexes such as MCP/MS2 RNA-stem-loop or CRISPR-Cas13/guide RNA (gRNA). MS2 RNA-stem-loops can be added to the RNA of interest so that the fusion protein containing the PL enzyme/MS2 protein (MCP- MS2 coat protein) recognizes the stem-loop(s) and sets the stage for PL (Yoon et al., 2012). CRISPR-Cas13/gRNA takes advantage of the endonuclease-deficient Cas13 (dCas13) and its ability to be directed to target RNA with a sequence-specific gRNA (Lin et al., 2021). As mentioned beforehand, the fusion protein would now be targeted to the RNA of interest and can start the PL process.
        Box 1. Controls and QC for RNA-centric probe pulldowns and proximity-labeling methods:
        For RNA-centric probe pulldowns, the negative controls must involve using beads-only or no probe control to verify that proteins are not binding non-specifically to the coated beads or matrix (Panda et al., 2016). Utilizing a scrambled non-targeting probe is required to ensure that the identified RBP is sequence or structure specific (Panda et al., 2016; Ramanathan et al., 2019; Torres et al., 2018). Furthermore, the use of different sets of probes against different regions of the targeted RNA helps evaluate the specificity and validate binding, while the pulldown targets themselves should be confirmed by using unrelated or endogenous RNAs as controls (Panda et al., 2016; Torres et al., 2018). Incorporating an RNase step that digests all unprotected RNA and yields the RNA bound to the RBP serves as an additional experimental control (Panda et al., 2016; Zhou et al., 2022).
        For proximity labeling methods, an experiment with an empty gRNA-expressing vector can help to define the baseline of background biotinylation (Yi et al., 2020). Other essential controls to evaluate include using a scrambled non-targeting guide RNA, an inactive enzyme, or adding no substrate for the enzyme to function (Giambruno & Nicassio, 2022; Jiang et al., 2025). Additionally, the labeling enzyme can be targeted to a different organelle or protein complex in addition to the primary target to verify localization specificity (Bosch et al., 2021). The checklist requires verifying other viral experimental controls, including non-crosslinked samples, non-immunoprecipitated samples, and lysates from RBP knockout samples (Sahadevan et al., 2026). Lastly, to properly validate the overall experimental setup and downstream proteomic analysis, RNA with known RBP partners must be included as positive controls, alongside validation by western blot and qRT-PCR (Panda et al., 2016; Ramanathan et al., 2019).

2.3. Other Experimental Methods to Identify RBPs

While often overlooked, there have been some advances in the development of in situ and ex vivo methods for identifying RBPs. These techniques are very diverse, each of them following its own principle. This creates a wide range of alternatives for identifying and cross-validating RBPs interacting with the target RNA, as shown in Table 4. For example, hybridization-proximity labeling (HyPro) relies on the hybridization of digoxigenin-labeled antisense probes to the RNA of interest. Then the HyPro enzyme, which is an APEX2/DIG10.3 fusion protein, binds to the digoxigenin probes. Any unbound enzyme is washed off, and the interactome of the RNA of interest is biotinylated in situ. After crosslink reversal, the labeled interacting molecules are captured on streptavidin beads and analyzed by MS. While similar to the in vivo PBL methods, HyPro-MS utilizes chemically fixed cells (instead of live cells) for the hybridization of the antisense probes to the RNA molecules(Yap et al., 2022).
Oligonucleotide-mediated proximity-interactome MAPping (O-MAP) combines RNA-FISH oligonucleotide probes and proximity biotinylating to identify RBPs. Oligonucleotide probes containing a universal “landing pad” sequence are annealed to the RNA of interest in fixed cells. A Horseradish Peroxidase (HRP) conjugated to a secondary oligo is then delivered to the universal sequence. Like other PBL enzymes, in the presence of biotin-tyramide and H2O2, HRP biotinylates molecules surrounding the transcript of interest. This method can identify various parts of the RNA’s interactome like other RNAs through sequencing (O-MAP-Seq), genomic loci (O-MAP-ChIP), and proteins (O-MAP-MS) (Tsue et al., 2024).
MS-based techniques have allowed characterization of protein folding and stability in crude biological samples like cell lysates. Examples of these methods involve thermal proteome profiling (TPP), stability of protein from rates of oxidation (SPROX), limited proteolysis (LiP), etc (Figure 6). Recently, some of these techniques have been adapted to elucidate the protein interactome of an RNA of interest, using unmodified RNA molecules to study RNA-protein interactions of various affinities. The principle behind these proteome-wide stability assays is the change in conformational and thermodynamic stability when an RNA ligand binds a protein, making the protein less susceptible to denaturation or proteolysis. This set of methods subject the lysate with/without the target RNA to denaturant conditions and proceed to analyze them via LC-MS/MS. For example, TPP utilizes a temperature gradient to denature aliquots of lysate proteins in the presence/absence of RNA ligand. The samples are then centrifuged, and the soluble protein fractions from the different aliquots are combined into a single tube to be analyzed via LC/MS-MS-based quantitative bottom-up proteomics. SPROX utilizes a urea concentration gradient as a denaturant. After exposing the aliquots of RNA/RNA-free protein lysate to urea, a methionine oxidation reaction is initiated by the addition of H2O2. Once the reaction is quenched, the gradient aliquots are combined into a single tube and prepared for LC-MS/MS quantitative bottom-up proteomics. By comparing the readout of the control (-) against the RNA-containing (+) samples, the proteins that are bound to RNA can be determined. Consequently, the identification of proteins that interact with the target RNA relies on shifts in protein stability/oxidation transitions. SPROX monitors protein oxidation rates via a peptide-level readout, whereas TPP relies on thermal profiling to provide a protein-level readout (Bailey et al., 2024). The identified RBPs can be validated by orthogonal methods such as pulldown and CLIP methods.

2.4. In Silico Methods to Identify RBPs

In silico approaches to identify RNA-binding proteins (RBPs) differ from traditional in vivo and in vitro methods in that they rely on computational data rather than experimental testing. These methods use algorithms that have been trained on experimentally validated data. A variety of methods have been developed following this template as seen in Table 5. An important advantage of in silico approaches is their ability to predict genome-wide RNA-RBP interaction candidates rapidly and cost-effectively. For example, bipartite motif finder (BMF) is the first thermodynamic tool used to identifying bipartite RNA motifs and model the binding affinity of RBPs, and it has demonstrated to detect short and degenerate motifs(Sohrabi-Jahromi & Söding, 2021).
One of the recent in silico tools used to detect these interactions is the fRNC (framework of RBP-ncRNA Circuits). This is a systems biology tool designed to identify condition-specific interaction circuits. These predictions are done by integrating transcriptomic, proteomic, and interatomic data. It constructs an RBP–ncRNA network using experimental datasets and can predict RBP-RNA networks in diseases with high confidence (Jiang et al., 2023).
Despite their advantages, existing in silico methods have limitations, including high dimensionality, data sparsity, and low model performance(Wang et al., 2022). Recently, motif discovery tools employing deep neural networks have been developed to predict RBP binding sites, by incorporating both sequence and secondary structure information. Deep convolutional neural networks are particularly effective for high-dimensional, sparse data(Sohrabi-Jahromi & Söding, 2021; Wang et al., 2022). However, as the number of model parameters and network complexity increase, so does the risk of prediction errors, especially since RBPs often bind to low-complexity, untranslated regions of RNA with lower binding affinities(Sohrabi-Jahromi & Söding, 2021).
To improve the performance, an evolutionary deep convolutional neural network (EDCNN) was developed to identify protein-RNA interactions by combining evolutionary algorithms and different gradient descent models(Wang et al., 2022). Traditional examples of databases include RBPDB and catRAPID, which are used to search for RNA sequences that match RBPs' binding preferences and to calculate the binding properties between proteins and RNA, respectively(Xu & Cui, 2025). A variety of newer datasets that build off of this concept are detailed in Table 6. RNAelem also uses deep neural networks and is designed to predict sequence-structure motifs in the binding regions of RBPs with high accuracy, as demonstrated using simulated data(Miyake et al., 2024).
Tools to predict circRNA-RBPs interactions: A graph neural network approach for predicting circRNA–RBP (GGCRB) model integrates nucleotide sequence information with RNA secondary features through graph convolution and attention mechanism to achieve a comprehensive characterization of multiscale features of circular RNAs. Overall, the GGCRB model offers a new alternative for circRNA-RBP prediction, capturing local and global features of circRNAs and achieving an AUC-ROC of 0.9634 (Tang et al., 2025).

3. Discussion

There have been several advancements in the characterization and identification of RNA-binding proteins, resulting in a wide range of available methods. Although these methodologies have expanded our knowledge on RNA-RBP interactions, limitations remain, and there is a particular need for user-friendly approaches to help researchers select appropriate technologies based on their specific needs. Selecting the optimal method for RNA-binding protein (RBP) identification can be challenging given the plethora of new methodologies. To address this, we have provided a practical decision-making guide (Figure 7) tailored to help researchers navigate current experimental and computational options based on experimental design. The first critical step is defining the biological model (immortalized versus primary cells) and the core context of the study. In vitro or ex vivo techniques are highly effective for global, system-level analyses, whereas in vivo and in situ methods are crucial for mapping physiologically relevant interactions in their native environments. Beyond these foundational decisions, the flowchart guides users through practical considerations that ultimately dictate method feasibility: reagent cost, required throughput, experimental turnaround time, and whether the chosen cell model can tolerate the genetic manipulations required by many proximity-labeling techniques.
A major advancement in this field would be the integration of bioinformatics tools and artificial intelligence with experimental methods. Increasing the selection of computational methods that are trained on the experimental data will allow for higher-quality predictions on binding specificity, RBP-RNA dependability, and co-binding partners. Computational methods will also enable a better understanding of how these interactions function in a disease setting. The future of RBP discovery lies in a mixed-methods approach, in which computational predictions are used to prioritize statistically significant candidate proteins and high-resolution experimental techniques are employed to validate these predictions. For example, deep learning models and RNA-protein interface databases have become greatly used in the last couple of years because they have improved approaches for predicting RNA-protein interactions and their binding sites and help narrow down list the of candidates (Konde et al., 2025; Miyake et al., 2024). However, there is still a need for experimental validation of these predictions.
Another area where current methods are lacking is the reliance on immunoprecipitation to isolate RBPs, usually done with an antibody or a capture probe. This dependency limits these methods and the overall discovery of RBPs to proteins for which a specific tag is available. Many unconventional RBPs, or RBPs that transiently interact with RNA, may be missed or poorly captured due to antibody non-specificity or failure. To unlock the potential of complete RBP discovery and identification, an approach centered on direct and quantitative physical binding, rather than antibody binding, would be a required innovation.
Overall, no single RBP identification technique can provide a complete picture of the interactome. Ideally, results from one method should be cross-validated with other RNA-based or protein-based approaches like the -seq techniques to ensure robust validation of findings.

Supplementary Materials

The following supporting information can be downloaded at Preprints.org, Table S1: Examples of known RNA-RBP interactions in diseases.

Author Contributions

Conceptualization, MB.; writing—original draft preparation, BR, JAFB, VB, and MH; writing—review and editing, BR, JAFB, VB, VP and MB; supervision, MB All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Science Foundation, grant number 2244127, to M.B. and V.P.

Institutional Review Board Statement

Not applicable.

Data Availability Statement

No new data were created or analyzed in this study.

Acknowledgments

We thank Aarush Sudamalla for helpful discussion during initial writing of this manuscript. During the writing and revisions of this work, the author(s) used Gemini to polish the English and grammar and used Biorender.ai to generate graphics for the figures. After using these tools/services, the authors have reviewed and edited the output and take full responsibility for the content of this publication.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Aeby, E., Lee, H. G., Lee, Y. W., Kriz, A., Del Rosario, B. C., Oh, H. J.,…Lee, J. T. (2020). Decapping enzyme 1A breaks X-chromosome symmetry by controlling Tsix elongation and RNA turnover. Nat Cell Biol, 22(9), 1116-1129. [CrossRef]
  2. Ahsan, F., Yan, Z., Precup, D., & Blanchette, M. (2022). PhyloPGM: boosting regulatory function prediction accuracy using evolutionary information. Bioinformatics, 38(Suppl 1), i299-i306. [CrossRef]
  3. Anastasakis, D., Benhalevy, D., & Hafner, M. (2020). Proximity-CLIP and Expedited Non-Radioactive Library Preparation of Small RNA Footprints for Next-Generation Sequencing. Curr Protoc Mol Biol, 131(1), e120. [CrossRef]
  4. Aydin, J., Gabel, A., Zielinski, S., Ganskih, S., Schmidt, N., Hartigan, C. R.,…Munschauer, M. (2024). SHIFTR enables the unbiased identification of proteins bound to specific RNA regions in live cells. Nucleic Acids Res, 52(5), e26. [CrossRef]
  5. Bailey, M. A., Martyr, J. G., Hargrove, A. E., & Fitzgerald, M. C. (2024). Stability-Based Proteomics for Investigation of Structured RNA-Protein Interactions. Anal Chem. [CrossRef]
  6. Benoit Bouvrette, L. P., Bovaird, S., Blanchette, M., & Lécuyer, E. (2020). oRNAment: a database of putative RNA binding protein target sites in the transcriptomes of model species. Nucleic Acids Res, 48(D1), D166-D173. [CrossRef]
  7. Benoit Bouvrette, L. P., Wang, X., Boulais, J., Kong, J., Syed, E. U., Blue, S. M.,…Lécuyer, E. (2023). RBP Image Database: A resource for the systematic characterization of the subcellular distribution properties of human RNA binding proteins. Nucleic Acids Res, 51(D1), D1549-D1557. [CrossRef]
  8. Biswas, J., Rahman, R., Gupta, V., Rosbash, M., & Singer, R. H. (2020). MS2-TRIBE Evaluates Both Protein-RNA Interactions and Nuclear Organization of Transcription by RNA Editing. iScience, 23(7), 101318. [CrossRef]
  9. Bosch, J. A., Chen, C. L., & Perrimon, N. (2021). Proximity-dependent labeling methods for proteomic profiling in living cells: An update. Wiley Interdiscip Rev Dev Biol, 10(1), e392. [CrossRef]
  10. Bourdon, S., Herviou, P., Dumas, L., Destefanis, E., Zen, A., Cammas, A.,…Dassi, E. (2023). QUADRatlas: the RNA G-quadruplex and RG4-binding proteins database. Nucleic Acids Res, 51(D1), D240-D247. [CrossRef]
  11. Cao, C., Wang, C., Dai, Q., Zou, Q., & Wang, T. (2024). CRBPSA: CircRNA-RBP interaction sites identification using sequence structural attention model. BMC Biol, 22(1), 260. [CrossRef]
  12. Cao, L., Wang, Y., Liu, J., Bai, X., & Chi, X. (2024). Long non-coding RNA TPT1-AS1 inhibits ferroptosis in ovarian cancer by regulating GPX4 via CREB1 regulation. Am J Reprod Immunol, 92(2), e13864. [CrossRef]
  13. Caudron-Herger, M., Jansen, R. E., Wassmer, E., & Diederichs, S. (2021). RBP2GO: a comprehensive pan-species database on RNA-binding proteins, their interactions and functions. Nucleic Acids Res, 49(D1), D425-D436. [CrossRef]
  14. Chen, H. C., Wang, J., Coffey, R. J., Patton, J. G., Weaver, A. M., Shyr, Y., & Liu, Q. (2024). EVPsort: An Atlas of Small ncRNA Profiling and Sorting in Extracellular Vesicles and Particles. J Mol Biol, 436(17), 168571. [CrossRef]
  15. Chu, H. P., Minajigi, A., Chen, Y., Morris, R., Guh, C. Y., Hsieh, Y. H.,…Lee, J. T. (2021). iDRiP for the systematic discovery of proteins bound directly to noncoding RNA. Nat Protoc, 16(7), 3672-3694. [CrossRef]
  16. Dal Molin, A., Gaffo, E., Difilippo, V., Buratin, A., Tretti Parenzan, C., Bresolin, S., & Bortoluzzi, S. (2022). CRAFT: a bioinformatics software for custom prediction of circular RNA functions. Brief Bioinform, 23(2). [CrossRef]
  17. Dodel, M., Guiducci, G., Dermit, M., Krishnamurthy, S., Alard, E. L., Capraro, F.,…Mardakheh, F. K. (2024). TREX reveals proteins that bind to specific RNA regions in living cells. Nat Methods, 21(3), 423-434. [CrossRef]
  18. Fiorentino, J., Armaos, A., Colantoni, A., & Tartaglia, G. G. (2024). Prediction of protein-RNA interactions from single-cell transcriptomic data. Nucleic Acids Res, 52(6), e31. [CrossRef]
  19. Fu, L., Zhang, D., Yi, N., Cao, Y., Wei, Y., Wang, W., & Li, L. (2022). Circular RNA circPBX3 promotes cisplatin resistance of ovarian cancer cells via interacting with IGF2BP2 to stabilize ATP7A mRNA expression. Hum Cell, 35(5), 1560-1576. [CrossRef]
  20. Gebauer, F., Schwarzl, T., Valcárcel, J., & Hentze, M. W. (2021). RNA-binding proteins in human genetic disease. Nat Rev Genet, 22(3), 185-198. [CrossRef]
  21. Ghanbari, M., & Ohler, U. (2020). Deep neural networks for interpreting RNA-binding protein target preferences. Genome Res, 30(2), 214-226. [CrossRef]
  22. Giambruno, R., & Nicassio, F. (2022). Proximity-dependent biotinylation technologies for mapping RNA-protein interactions in live cells. Front Mol Biosci, 9, 1062448. [CrossRef]
  23. Gresova, K., Racek, T., Martinek, V., Cechak, D., Svobodova, R., & Alexiou, P. (2023). RBP-Tar - a searchable database for experimental RBP binding sites. F1000Res, 12, 755. [CrossRef]
  24. Guo, J., Guo, S., Lu, S., Gong, J., Wang, L., Ding, L.,…Liu, W. (2023). The development of proximity labeling technology and its applications in mammals, plants, and microorganisms. Cell Commun Signal, 21(1), 269. [CrossRef]
  25. Guo, Y., Lei, X., & Li, S. (2025). An Integrated TCN-CrossMHA Model for Predicting circRNA-RBP Binding Sites. Interdiscip Sci, 17(1), 86-100. [CrossRef]
  26. Han, S., Zhao, B. S., Myers, S. A., Carr, S. A., He, C., & Ting, A. Y. (2020). RNA-protein interaction mapping via MS2- or Cas13-based APEX targeting. Proc Natl Acad Sci U S A, 117(36), 22068-22079. [CrossRef]
  27. Hao, L., Zhang, J., Liu, Z., Zhang, Z., Mao, T., & Guo, J. (2023). Role of the RNA-binding protein family in gynecologic cancers. Am J Cancer Res, 13(8), 3799-3821.
  28. Hitachi, K., Kiyofuji, Y., Nakatani, M., & Tsuchida, K. (2021). -Associated and -Independent Multiple Roles of Heterogeneous Nuclear Ribonucleoprotein K during Skeletal Muscle Cell Differentiation. Int J Mol Sci, 23(1). [CrossRef]
  29. Huang, Y., Zhang, L., Mu, W., Zheng, M., Bao, X., Li, H.,…Zuo, Z. (2025). RMVar 2.0: an updated database of functional variants in RNA modifications. Nucleic Acids Res, 53(D1), D275-D283. [CrossRef]
  30. Hynes, C., & Kakumani, P. K. (2024). Regulatory role of RNA-binding proteins in microRNA biogenesis. Front Mol Biosci, 11, 1374843. [CrossRef]
  31. Ikegawa, M., Kano, N., Ori, D., Fukuta, M., Hirano, M., Hewson, R.,…Kawasaki, T. (2024). HuR (ELAVL1) regulates the CCHFV minigenome and HAZV replication by associating with viral genomic RNA. PLoS Negl Trop Dis, 18(9), e0012553. [CrossRef]
  32. Jansen, S., Smlatic, E., Copmans, D., Debaveye, S., Tangy, F., Vidalain, P. O.,…Dallmeier, K. (2021). Identification of host factors binding to dengue and Zika virus subgenomic RNA by efficient yeast three-hybrid screens of the human ORFeome. RNA Biol, 18(5), 732-744. [CrossRef]
  33. Jasinski-Bergner, S., Vaxevanis, C., Heimer, N., Lazaridou, M. F., Friedrich, M., & Seliger, B. (2020). An altered miTRAP method for miRNA affinity purification with its pros and cons. Methods Enzymol, 636, 323-337. [CrossRef]
  34. Jeffrey, D. A., Dockstader, K., Revoredo Vicentino, A., Slavov, D. B., Miyamoto, S. D., Stauffer, B. L., & Sucharov, C. C. (2025). Muscle-specific isoforms of FXR1 are necessary for miR-1-mediated repression of connexin 43 and are downregulated in pediatric dilated cardiomyopathy. Am J Physiol Heart Circ Physiol, 328(6), H1380-H1390. [CrossRef]
  35. Jiang, F., Ge, X., & Bennett, E. J. (2025). Employing Expression-Matched Controls Enables High-Confidence Proximity-Based Interactome Classification. Mol Cell Proteomics, 24(7), 101001. [CrossRef]
  36. Jiang, L., Hao, S., Lin, L., Gao, X., & Xu, J. (2023). RNC: Uncovering the dynamic and condition-specific RBP-ncRNA circuits from multi-omics data. Comput Struct Biotechnol J, 21, 2276-2285. [CrossRef]
  37. Kalocsay, M. (2019). APEX Peroxidase-Catalyzed Proximity Labeling and Multiplexed Quantitative Proteomics. Methods Mol Biol, 2008, 41-55. [CrossRef]
  38. Kelaini, S., Chan, C., Cornelius, V. A., & Margariti, A. (2021). RNA-Binding Proteins Hold Key Roles in Function, Dysfunction, and Disease. Biology (Basel), 10(5). [CrossRef]
  39. Kim, J., & Ma, L. (2021). Capturing Endogenous Long Noncoding RNAs and Their Binding Proteins Using Chromatin Isolation by RNA Purification. Methods Mol Biol, 2372, 85-92. [CrossRef]
  40. Kodavati, M., Maloji Rao, V. H., Provasek, V. E., & Hegde, M. L. (2024). Regulation of DNA damage response by RNA/DNA-binding proteins: Implications for neurological disorders and aging. Ageing Res Rev, 100, 102413. [CrossRef]
  41. Konde, O. M., Balela, W. B., & Tshitenge, T. B. (2025). Targeting RNA-binding proteins with small molecules: perspectives and challenges. Front Chem, 13, 1649692. [CrossRef]
  42. Kristofich, J., & Nicchitta, C. V. (2024). Rapid and Efficient Isolation of Total RNA-Bound Proteomes by Liquid Emulsion-Assisted Purification of RNA-Bound Protein (LEAP-RBP). Bio Protoc, 14(14), e5236. [CrossRef]
  43. Lasantha, D., Vidanagamachchi, S., & Nallaperuma, S. (2024). CRIECNN: Ensemble convolutional neural network and advanced feature extraction methods for the precise forecasting of circRNA-RBP binding sites. Comput Biol Med, 174, 108466. [CrossRef]
  44. Li, H., Deng, Z., Yang, H., Pan, X., Wei, Z., Shen, H. B.,…Wu, J. (2022). circRNA-binding protein site prediction based on multi-view deep learning, subspace learning and multi-view classifier. Brief Bioinform, 23(1). [CrossRef]
  45. Li, K., Guo, Z. W., Zhai, X. M., Yang, X. X., Wu, Y. S., & Liu, T. C. (2020). RBPTD: a database of cancer-related RNA-binding proteins in humans. Database (Oxford), 2020. [CrossRef]
  46. Li, Y., Liu, S., Cao, L., Luo, Y., Du, H., Li, S.,…You, F. (2021). CBRPP: a new RNA-centric method to study RNA-protein interactions. RNA Biol, 18(11), 1608-1621. [CrossRef]
  47. Liang, Q., Yu, T., Kofman, E., Jagannatha, P., Rhine, K., Yee, B. A.,…Yeo, G. W. (2024). High-sensitivity in situ capture of endogenous RNA-protein interactions in fixed cells and primary tissues. Nat Commun, 15(1), 7067. [CrossRef]
  48. Liao, J. Y., Yang, B., Zhang, Y. C., Wang, X. J., Ye, Y., Peng, J. W.,…Yin, D. (2020). EuRBPDB: a comprehensive resource for annotation, functional and oncological investigation of eukaryotic RNA binding proteins (RBPs). Nucleic Acids Res, 48(D1), D307-D313. [CrossRef]
  49. Lin, X., Fonseca, M. A. S., Breunig, J. J., Corona, R. I., & Lawrenson, K. (2021). discovery of RNA proximal proteins via proximity-dependent biotinylation. RNA Biol, 18(12), 2203-2217. [CrossRef]
  50. Lin, Y., Li, Y., Chen, Y., & Zhang, Z. (2022). LncRNA ALMS1-IT1 is a novel prognostic biomarker and correlated with immune infiltrates in colon adenocarcinoma. Medicine (Baltimore), 101(42), e31314. [CrossRef]
  51. Lindsey, G. L., Hockley, T. K., Gomez, A. V., Marshall, A. C., Brothers, W. R., Finney, C. T.,…Cravatt, B. F. (2026). Structural and mechanistic analysis of covalent ligands targeting the RNA-binding protein NONO. Cell Chem Biol. [CrossRef]
  52. Liu, L., Wei, Y., Tan, Z., Zhang, Q., Sun, J., & Zhao, Q. (2024). Predicting circRNA-RBP Binding Sites Using a Hybrid Deep Neural Network. Interdiscip Sci, 16(3), 635-648. [CrossRef]
  53. Liu, L., Wei, Y., Zhang, Q., & Zhao, Q. (2024). SSCRB: Predicting circRNA-RBP Interaction Sites Using a Sequence and Structural Feature-Based Attention Model. IEEE J Biomed Health Inform, 28(3), 1762-1772. [CrossRef]
  54. Liu, N., & Zhang, Y. (2025). A deep learning approach based on molecular graph features and residual blocks to predict interaction sites between CircRNA and RBP. Biochem Biophys Res Commun, 785, 152655. [CrossRef]
  55. Liu, W., Wang, W., Wang, Z., Fan, X., Li, W., Huang, Y.,…Tang, Z. (2024). CRISPR Screen Identifies the RNA-Binding Protein Eef1a1 as a Key Regulator of Myogenesis. Int J Mol Sci, 25(9). [CrossRef]
  56. Marcelo, A., Koppenol, R., de Almeida, L. P., Matos, C. A., & Nóbrega, C. (2021). Stress granules, RNA-binding proteins and polyglutamine diseases: too much aggregation? Cell Death Dis, 12(6), 592. [CrossRef]
  57. Miyake, H., Kawaguchi, R. K., & Kiryu, H. (2024). RNAelem: an algorithm for discovering sequence-structure motifs in RNA bound by RNA-binding proteins. Bioinform Adv, 4(1), vbae144. [CrossRef]
  58. Mu, K., Fei, Y., Xu, Y., & Zhang, Q. C. (2025). RASP v2.0: an updated atlas for RNA structure probing data. Nucleic Acids Res, 53(D1), D211-D219. [CrossRef]
  59. Nabeel-Shah, S., & Greenblatt, J. F. (2023). Revised iCLIP-seq Protocol for Profiling RNA-protein Interaction Sites at Individual Nucleotide Resolution in Living Cells. Bio Protoc, 13(11), e4688. [CrossRef]
  60. Nicholson-Shaw, A. L., Kofman, E. R., Yeo, G. W., & Pasquinelli, A. E. (2022). Nuclear and cytoplasmic poly(A) binding proteins (PABPs) favor distinct transcripts and isoforms. Nucleic Acids Res, 50(8), 4685-4702. [CrossRef]
  61. Oberstrass, L., Tants, J. N., Lichtenthaeler, C., Ali, S. E., Koch, L., Mathews, D. H.,…Weigand, J. E. (2024). Comprehensive Profiling of Roquin Binding Preferences for RNA Stem-Loops. Angew Chem Int Ed Engl, 63(50), e202412596. [CrossRef]
  62. Pan, X., Fang, Y., Liu, X., Guo, X., & Shen, H. B. (2025). RBPsuite 2.0: an updated RNA-protein binding site prediction suite with high coverage on species and proteins based on deep learning. BMC Biol, 23(1), 74. [CrossRef]
  63. Panda, A. C., Martindale, J. L., & Gorospe, M. (2016). Affinity Pulldown of Biotinylated RNA for Detection of Protein-RNA Complexes. Bio Protoc, 6(24). [CrossRef]
  64. Pant, P., & Kumarswamy, R. (2024). Multiple Oligo assisted RNA Pulldown via Hybridization followed by Mass Spectrometry (MORPH-MS) for exploring the RNA-Protein interactions. RNA Biol, 21(1), 1-9. [CrossRef]
  65. Qi, Z., Xue, S., Chen, J., Zhao, W., Johnson, K., Wen, X.,…Zhong, S. (2025). Genome-wide mapping of RNA-protein associations through sequencing. Nat Biotechnol. [CrossRef]
  66. Qiu, Y., Chen, W., Ching, W. K., Cai, H., Jiang, H., & Zou, Q. (2024). AGML: Adaptive Graph-Based Multi-Label Learning for Prediction of RBP and as Event Associations During EMT. IEEE/ACM Trans Comput Biol Bioinform, 21(6), 2113-2122. [CrossRef]
  67. Qiu, Y., Ching, W. K., & Zou, Q. (2021a). Matrix factorization-based data fusion for the prediction of RNA-binding proteins and alternative splicing event associations during epithelial-mesenchymal transition. Brief Bioinform, 22(6). [CrossRef]
  68. Qiu, Y., Ching, W. K., & Zou, Q. (2021b). Prediction of RNA-binding protein and alternative splicing event associations during epithelial-mesenchymal transition based on inductive matrix completion. Brief Bioinform, 22(5). [CrossRef]
  69. Ramanathan, M., Porter, D. F., & Khavari, P. A. (2019). Methods to study RNA-protein interactions. Nat Methods, 16(3), 225-234. [CrossRef]
  70. Sahadevan, S., Sekaran, T., Hentze, M. W., Perez-Perri, J. I., & Schwarzl, T. (2026). Signal or noise? RNA-binding proteins and the challenges of binding site assignments. Nucleic Acids Res, 54(9). [CrossRef]
  71. Savulescu, A. F., Stoychev, S., Mamputha, S., & Mhlanga, M. M. (2020). Biochemical Pulldown of mRNAs and Long Noncoding RNAs from Cellular Lysates Coupled with Mass Spectrometry to Identify Protein Binding Partners. Bio Protoc, 10(11), e3639. [CrossRef]
  72. Sen, S., & Mukhopadhyay, D. (2024). A Holistic Analysis of Alzheimer's Disease-Associated lncRNA Communities Reveals Enhanced lncRNA-miRNA-RBP Regulatory Triad Formation Within Functionally Segregated Clusters. J Mol Neurosci, 74(3), 77. [CrossRef]
  73. Shen, Y., Chen, H., Huang, Q., Du, H., & Zhou, L. (2022). Transcriptomic signature associated with RNA-binding proteins for survival stratification of laryngeal cancer. Aging (Albany NY), 14(16), 6605-6625. [CrossRef]
  74. Sohrabi-Jahromi, S., & Söding, J. (2021). Thermodynamic modeling reveals widespread multivalent binding by RNA-binding proteins. Bioinformatics, 37(Suppl_1), i308-i316. [CrossRef]
  75. Sommerauer, C., & Kutter, C. (2022). Noncoding RNAs and RNA-binding proteins: emerging governors of liver physiology and metabolic diseases. Am J Physiol Cell Physiol, 323(4), C1003-C1017. [CrossRef]
  76. Srivastava, M., Srivastava, R., & Janga, S. C. (2021). Transcriptome-wide high-throughput mapping of protein-RNA occupancy profiles using POP-seq. Sci Rep, 11(1), 1175. [CrossRef]
  77. Street, L. A., Rothamel, K. L., Brannan, K. W., Jin, W., Bokor, B. J., Dong, K.,…Yeo, G. W. (2024). Large-scale map of RNA-binding protein interactomes across the mRNA life cycle. Mol Cell, 84(19), 3790-3809.e3798. [CrossRef]
  78. Sun, L., Xu, K., Huang, W., Yang, Y. T., Li, P., Tang, L.,…Zhang, Q. C. (2021). Predicting dynamic cellular protein-RNA interactions by deep learning using in vivo RNA structures. Cell Res, 31(5), 495-516. [CrossRef]
  79. Sun, X., Li, W., Li, G., Yang, H., Jiang, Z., Shen, L.,…Wang, G. (2024). A novel long non-coding RNA LINC00524 facilitates invasion and metastasis through interaction with TDP43 in breast cancer. J Cell Mol Med, 28(8), e18275. [CrossRef]
  80. Tang, G., Xing, H., Yao, D., Zhan, X., & Li, X. (2025). GGCRB: A Graph Neural Network Approach for Predicting CircRNA-RBP Interactions Using Structural and Sequence Features. ACS Omega, 10(30), 33662-33674. [CrossRef]
  81. Tankka, A. T., Zhang, Y., Einstein, J. M., Zhou, C. J., Pham, V. N., Naritomi, J. T.,…Yeo, G. W. (2025). Integrative CRISPR Screening and RNA Analyses Discover an Essential Role for PUF60 Interactions with 3' Splice Sites in Cancer Progression. Cancer Res. [CrossRef]
  82. Teng, X., Shang, J., Du, L., Huang, W., Wang, Y., Liu, M.,…Bai, L. (2024). RNA-binding protein Trx regulates alternative splicing and promotes metastasis of HCC via interacting with LINC00152. J Gastroenterol Hepatol, 39(12), 2892-2902. [CrossRef]
  83. Torres, M., Becquet, D., Guillen, S., Boyer, B., Moreno, M., Blanchard, M. P.,…François-Bellan, A. M. (2018). RNA Pull-down Procedure to Identify RNA Targets of a Long Non-coding RNA. J Vis Exp(134). [CrossRef]
  84. Tsue, A. F., Kania, E. E., Lei, D. Q., Fields, R., McGann, C. D., Marciniak, D. M.,…Shechner, D. M. (2024). Multiomic characterization of RNA microenvironments by oligonucleotide-mediated proximity-interactome mapping. Nat Methods, 21(11), 2058-2071. [CrossRef]
  85. Vemuri, S., Srivastava, R., Mir, Q., Hashemikhabir, S., Dong, X. C., & Janga, S. C. (2020). SliceIt: A genome-wide resource and visualization tool to design CRISPR/Cas9 screens for editing protein-RNA interaction sites in the human genome. Methods, 178, 104-113. [CrossRef]
  86. Wang, J., Horlacher, M., Cheng, L., & Winther, O. (2024). DeepLocRNA: an interpretable deep learning model for predicting RNA subcellular localization with domain-specific transfer-learning. Bioinformatics, 40(2). [CrossRef]
  87. Wang, Y., Yang, Y., Ma, Z., Wong, K. C., & Li, X. (2022). EDCNN: identification of genome-wide RNA-binding proteins using evolutionary deep convolutional neural network. Bioinformatics, 38(3), 678-686. [CrossRef]
  88. Wang, Z., & Lei, X. (2020). Matrix factorization with neural network for predicting circRNA-RBP interactions. BMC Bioinformatics, 21(1), 229. [CrossRef]
  89. Wang, Z., & Lei, X. (2021). Identifying the sequence specificities of circRNA-binding proteins based on a capsule network architecture. BMC Bioinformatics, 22(1), 19. [CrossRef]
  90. Wei, Y., Tan, Z., & Liu, L. (2025). CR-deal: Explainable Neural Network for circRNA-RBP Binding Site Recognition and Interpretation. Interdiscip Sci, 17(2), 463-476. [CrossRef]
  91. Weissinger, R., Heinold, L., Akram, S., Jansen, R. P., & Hermesh, O. (2021). RNA Proximity Labeling: A New Detection Tool for RNA-Protein Interactions. Molecules, 26(8). [CrossRef]
  92. Wu, S., & Guo, J. T. (2024). Improved prediction of DNA and RNA binding proteins with deep learning models. Brief Bioinform, 25(4). [CrossRef]
  93. Xing, S., Wang, J., Wu, R., Hefti, M. M., Crary, J. F., & Lu, Y. (2022). Identification of HnRNPC as a novel Tau exon 10 splicing factor using RNA antisense purification mass spectrometry. RNA Biol, 19(1), 104-116. [CrossRef]
  94. Xiong, Y., Wang, A., Kang, Y., Shen, C., Hsieh, C. Y., & Hou, T. (2025). mRNABERT: advancing mRNA sequence design with a universal language model and comprehensive dataset. Nat Commun, 16(1), 10371. [CrossRef]
  95. Xu, Z., & Cui, Y. (2025). The dual role of RNA-binding proteins: promotion of tumorigenesis, drug resistance, and emerging therapeutic targets. Acta Biochim Biophys Sin (Shanghai), 57(12), 1923-1938. [CrossRef]
  96. Yan, A., Xiong, J., Zhu, J., Li, X., Xu, S., Feng, X.,…Kee, K. (2022). DAZL regulates proliferation of human primordial germ cells by direct binding to precursor miRNAs and enhances DICER processing activity. Nucleic Acids Res, 50(19), 11255-11272. [CrossRef]
  97. Yang, D., Li, C., Kong, Y., Pei, Y., Miao, B., Dai, G.,…Pei, R. (2024). Deciphering the Temporal-Spatial Interactive Heterogeneity of Long Non-Coding RNAs and RNA-Binding Proteins in Living Cells at Single-Cell Resolution. J Am Chem Soc, 146(30), 20878-20890. [CrossRef]
  98. Yang, J., Liu, X., Zhao, Y., Dong, W., Xue, Y., Ruan, X.,…Liu, Y. (2023). Mechanism of Dcp2/RNCR3/Dkc1/Snora62 axis regulating neuronal apoptosis in chronic cerebral ischemia. Cell Biol Toxicol, 39(6), 2881-2898. [CrossRef]
  99. Yang, X., Han, B., Zhang, R., Su, Y., Hosseini, D. K., Wu, H.,…Sun, H. (2021). Development and validation of a RNA binding protein-associated prognostic model for head and neck squamous cell carcinoma. Aging (Albany NY), 13(6), 7975-7997. [CrossRef]
  100. Yang, Z., Wan, J., Ma, L., Li, Z., Yang, R., Yang, H.,…Ming, L. (2023). Long non-coding RNA HOXC-AS1 exerts its oncogenic effects in esophageal squamous cell carcinoma by interaction with IGF2BP2 to stabilize SIRT1 expression. J Clin Lab Anal, 37(1), e24801. [CrossRef]
  101. Yap, K., Chung, T. H., & Makeyev, E. V. (2022). Hybridization-proximity labeling reveals spatially ordered interactions of nuclear RNA compartments. Mol Cell, 82(2), 463-478.e411. [CrossRef]
  102. Yi, W., Li, J., Zhu, X., Wang, X., Fan, L., Sun, W.,…Yan, J. (2020). CRISPR-assisted detection of RNA-protein interactions in living cells. Nat Methods, 17(7), 685-688. [CrossRef]
  103. Yin, W., Lei, Y., Yang, X., & Zou, J. (2023). A two-gene random forest model to diagnose osteoarthritis based on RNA-binding protein-related genes in knee cartilage tissue. Aging (Albany NY), 15(1), 193-212. [CrossRef]
  104. Yoon, J. H., Srikantan, S., & Gorospe, M. (2012). MS2-TRAP (MS2-tagged RNA affinity purification): tagging RNA to identify associated miRNAs. Methods, 58(2), 81-87. [CrossRef]
  105. Yu, T., Liang, Q., Xu, S., & Yeo, G. W. (2025). Identification of RBP binding sites using RNA deaminases. Methods Enzymol, 713, 287-297. [CrossRef]
  106. Zhang, J., Peng, Y., Cui, F., Zhang, Z., Yan, S., & Zhang, Q. (2025). RMDNet: RNA-aware dung beetle optimization-based multi-branch integration network for RNA-protein binding sites prediction. BMC Bioinformatics, 26(1), 176. [CrossRef]
  107. Zhang, Y., Liu, K., Xu, Z., Li, B., Wu, X., Fan, R.,…Xu, H. (2025). OncoSplicing 3.0: an updated database for identifying RBPs regulating alternative splicing events in cancers. Nucleic Acids Res, 53(D1), D1460-D1466. [CrossRef]
  108. Zhang, Z., Sun, W., Shi, T., Lu, P., Zhuang, M., & Liu, J. L. (2020). Capturing RNA-protein interaction via CRUIS. Nucleic Acids Res, 48(9), e52. [CrossRef]
  109. Zhao, W., Zhang, S., Zhu, Y., Xi, X., Bao, P., Ma, Z.,…Lu, Z. J. (2022). POSTAR3: an updated platform for exploring post-transcriptional regulation coordinated by RNA-binding proteins. Nucleic Acids Res, 50(D1), D287-D294. [CrossRef]
  110. Zheng, Y., Luo, H., Teng, X., Hao, X., Yan, X., Tang, Y.,…He, S. (2023). NPInter v5.0: ncRNA interaction database in a new era. Nucleic Acids Res, 51(D1), D232-D239. [CrossRef]
  111. Zhong, Y., Wilkinson-White, L., Zhang, E., Mohanty, B., Zhang, B. B., McRae, M. S.,…Kwan, A. H. (2024). Peptide nucleic acids can form hairpins and bind RNA-binding proteins. PLoS One, 19(9), e0310565. [CrossRef]
  112. Zhou, J., Zhao, H., Guo, Y., Hou, K., Ding, Q., Shao, W., & Xu, Q. (2023). Comprehensive Analysis of RNA-Binding Protein-Related lncRNA in Breast Invasive Carcinoma. Front Biosci (Landmark Ed), 28(1), 9. [CrossRef]
  113. Zhou, Y., Cui, H., Liu, D., & Wang, W. (2024). MSTCRB: Predicting circRNA-RBP interaction by extracting multi-scale features based on transformer and attention mechanism. Int J Biol Macromol, 278(Pt 2), 134805. [CrossRef]
  114. Zhou, Y., Sotcheff, S. L., & Routh, A. L. (2022). Next-generation sequencing: A new avenue to understand viral RNA-protein interactions. J Biol Chem, 298(5), 101924. [CrossRef]
  115. Zhu, H., Yang, Y., Wang, Y., Wang, F., Huang, Y., Chang, Y.,…Li, X. (2023). Dynamic characterization and interpretation for protein-RNA interactions across diverse cellular conditions using HDRNet. Nat Commun, 14(1), 6824. [CrossRef]
Preprints 223642 g001
Figure 1. Potential Functions of RNA-Binding Proteins (RBPs) in RNA life cycle. RBPs modulate the RNA life cycle through diverse mechanisms, including (A) the regulation of pre-miRNA biogenesis by proteins such as Drosha and Dicer(Hynes & Kakumani, 2024); (B) the mediation of mRNA localization by guiding transcripts to specific cellular destinations through the binding of RBPs to signaling sequences (Kelaini et al., 2021); and (C) the recruitment of RNases to short hairpin sequences to trigger mRNA degradation (Kelaini et al., 2021). Conversely, RBPs may also promote (D) mRNA stabilization by binding to the poly(A) tail, thereby shielding it from RNase activity (Kelaini et al., 2021). Furthermore, RBPs regulate translation by (E) acting as scaffolds for stress granule formation, thereby inducing translational stalling (Marcelo et al., 2021), and (F) influencing alternative splicing by targeting the inclusion or exclusion of exons (Gebauer et al., 2021). The RBPs shown here depict both direct binding and complex mediated mechanisms for simplicity. Created with BioRender.com.
Figure 1. Potential Functions of RNA-Binding Proteins (RBPs) in RNA life cycle. RBPs modulate the RNA life cycle through diverse mechanisms, including (A) the regulation of pre-miRNA biogenesis by proteins such as Drosha and Dicer(Hynes & Kakumani, 2024); (B) the mediation of mRNA localization by guiding transcripts to specific cellular destinations through the binding of RBPs to signaling sequences (Kelaini et al., 2021); and (C) the recruitment of RNases to short hairpin sequences to trigger mRNA degradation (Kelaini et al., 2021). Conversely, RBPs may also promote (D) mRNA stabilization by binding to the poly(A) tail, thereby shielding it from RNase activity (Kelaini et al., 2021). Furthermore, RBPs regulate translation by (E) acting as scaffolds for stress granule formation, thereby inducing translational stalling (Marcelo et al., 2021), and (F) influencing alternative splicing by targeting the inclusion or exclusion of exons (Gebauer et al., 2021). The RBPs shown here depict both direct binding and complex mediated mechanisms for simplicity. Created with BioRender.com.
Preprints 223642 g001
Preprints 223642 g002
Figure 2. General overview of in vitro and in vivo methods. (a) In vitro methods for identifying RNA-protein interactions often rely on RNA pull-down assays. (a.1-3) In vitro-transcribed, biotinylated RNA is coupled to streptavidin beads and incubated with cell lysates. Beads are washed and boiled to elute-RNA protein complexes for downstream analysis. (b) In vivo methods, on the other hand, rely on (b.1) cross-linking based methods, (b.2) proximity labeling-based techniques, and (b.3) enzymatic editing methods. In cross-linking-based approaches, focusing on RNA-centric methods (b1.1-4a), RNA-RBPs complexes are cross-linked either with UV light or with formaldehyde, captured using biotinylated antisense probes, and purified with streptavidin beads for analysis via mass spectrometry. Conversely, in protein-centric methods based on crosslinking (b.1.1-4b), the RBP-RNA complexes are crosslinked, a selected RBP is immunoprecipitated, and the bound RNA is identified via sequencing. Regarding proximity-labeling-based (PL) techniques (b2.1-4), a fusion protein targets specific RNA. Upon adding a biotin substrate (or an equivalent), the PL enzyme labels nearby molecules, which are then captured by streptavidin beads and eluted for analysis. Finally, enzymatic editing methods (b3.1-3) such as TRIBE function by fusing the RBP of interest to the enzyme ADAR (Adenosine Deaminase Acting on RNA), which deaminates nearby adenosines, and these altered bases are later detected using (Ramanathan et al., 2019). Created with BioRender.com.
Figure 2. General overview of in vitro and in vivo methods. (a) In vitro methods for identifying RNA-protein interactions often rely on RNA pull-down assays. (a.1-3) In vitro-transcribed, biotinylated RNA is coupled to streptavidin beads and incubated with cell lysates. Beads are washed and boiled to elute-RNA protein complexes for downstream analysis. (b) In vivo methods, on the other hand, rely on (b.1) cross-linking based methods, (b.2) proximity labeling-based techniques, and (b.3) enzymatic editing methods. In cross-linking-based approaches, focusing on RNA-centric methods (b1.1-4a), RNA-RBPs complexes are cross-linked either with UV light or with formaldehyde, captured using biotinylated antisense probes, and purified with streptavidin beads for analysis via mass spectrometry. Conversely, in protein-centric methods based on crosslinking (b.1.1-4b), the RBP-RNA complexes are crosslinked, a selected RBP is immunoprecipitated, and the bound RNA is identified via sequencing. Regarding proximity-labeling-based (PL) techniques (b2.1-4), a fusion protein targets specific RNA. Upon adding a biotin substrate (or an equivalent), the PL enzyme labels nearby molecules, which are then captured by streptavidin beads and eluted for analysis. Finally, enzymatic editing methods (b3.1-3) such as TRIBE function by fusing the RBP of interest to the enzyme ADAR (Adenosine Deaminase Acting on RNA), which deaminates nearby adenosines, and these altered bases are later detected using (Ramanathan et al., 2019). Created with BioRender.com.
Preprints 223642 g002
Preprints 223642 g003
Figure 3. Experimental workflow of PRIM-seq method. (a) Total RNA is isolated from cultured cells, and (b) a puromycin-conjugated linker is attached to the 3’ ends to enable in vitro translation of covalently linked mRNA-protein fusions. (c) The mRNA barcodes are reverse-transcribed into cDNA, yielding a library of cDNA-labeled proteins. (d) This library is incubated with a target RNA library to allow for RNA-protein interactions, followed by ligation to generate chimeric molecules. (e) The resulting chimeric products are sequenced, enabling the identification of RNA-RBP interactions (Qi et al., 2025). Created in BioRender.com.
Figure 3. Experimental workflow of PRIM-seq method. (a) Total RNA is isolated from cultured cells, and (b) a puromycin-conjugated linker is attached to the 3’ ends to enable in vitro translation of covalently linked mRNA-protein fusions. (c) The mRNA barcodes are reverse-transcribed into cDNA, yielding a library of cDNA-labeled proteins. (d) This library is incubated with a target RNA library to allow for RNA-protein interactions, followed by ligation to generate chimeric molecules. (e) The resulting chimeric products are sequenced, enabling the identification of RNA-RBP interactions (Qi et al., 2025). Created in BioRender.com.
Preprints 223642 g003
Preprints 223642 g004
Figure 4. Experimental protocol for MORPH-MS. (a) Cells used for each pull-down, including the odd probe, even probe, and RNase A-treated groups, are crosslinked in 3% paraformaldehyde in PBS and then subjected to chromatin shearing. (b)The complementary probes, divided into odd and even groups, are annealed with a universal oligo, and subsequently, the annealed probes are added to the samples to hybridize. (c) The pulldown is performed using streptavidin-coated magnetic beads. (d) After the beads are washed, the proteins are eluted with benzonase and identified by mass spectrometry (Pant & Kumarswamy, 2024). Created in BioRender.com.
Figure 4. Experimental protocol for MORPH-MS. (a) Cells used for each pull-down, including the odd probe, even probe, and RNase A-treated groups, are crosslinked in 3% paraformaldehyde in PBS and then subjected to chromatin shearing. (b)The complementary probes, divided into odd and even groups, are annealed with a universal oligo, and subsequently, the annealed probes are added to the samples to hybridize. (c) The pulldown is performed using streptavidin-coated magnetic beads. (d) After the beads are washed, the proteins are eluted with benzonase and identified by mass spectrometry (Pant & Kumarswamy, 2024). Created in BioRender.com.
Preprints 223642 g004
Preprints 223642 g005
Figure 5. General workflow of Proximity-Based Labeling. (a) The fusion protein targets the RNA of interest. (b) After a biotin substrate (or any substrate specific to the PL enzyme) is added, the PL enzyme biotinylates the surrounding molecules (including RBPs) within its labeling radius. (c) Subsequent to cell lysis, the lysate is incubated with streptavidin beads to enrich for biotinylated molecules. (d) Following incubation, the enriched molecules are analyzed by MS (e.g. LC-MS/MS) to elucidate their identity. Created with BioRender.com.
Figure 5. General workflow of Proximity-Based Labeling. (a) The fusion protein targets the RNA of interest. (b) After a biotin substrate (or any substrate specific to the PL enzyme) is added, the PL enzyme biotinylates the surrounding molecules (including RBPs) within its labeling radius. (c) Subsequent to cell lysis, the lysate is incubated with streptavidin beads to enrich for biotinylated molecules. (d) Following incubation, the enriched molecules are analyzed by MS (e.g. LC-MS/MS) to elucidate their identity. Created with BioRender.com.
Preprints 223642 g005
Preprints 223642 g006aPreprints 223642 g006b
Figure 6. General workflow of Proteome-Wide Stability Assays. (A) TPP workflow: aliquots are subjected to a thermal gradient, and after equilibration and cooling, the samples are pooled and centrifuged. The soluble protein supernatant is then collected for MS analysis (B) SPROX workflow: aliquots undergo a urea gradient, and after equilibration, samples undergo methionine, are quenched with TCEP, and pooled. Methionine-containing peptides are enriched using a Pi3 kit prior to MS analysis. Created with BioRender.com(Bailey et al., 2024).
Figure 6. General workflow of Proteome-Wide Stability Assays. (A) TPP workflow: aliquots are subjected to a thermal gradient, and after equilibration and cooling, the samples are pooled and centrifuged. The soluble protein supernatant is then collected for MS analysis (B) SPROX workflow: aliquots undergo a urea gradient, and after equilibration, samples undergo methionine, are quenched with TCEP, and pooled. Methionine-containing peptides are enriched using a Pi3 kit prior to MS analysis. Created with BioRender.com(Bailey et al., 2024).
Preprints 223642 g006aPreprints 223642 g006b
Preprints 223642 g007
Figure 7. Decision guide for selecting an RNA-binding protein (RBP) identification method. To use this chart, start at the far left by defining your primary desired output: mapping the global interactome, focusing on a specific RNA target, or investigating a specific protein. Follow the branches based on your subsequent experimental constraints—such as whether you require direct binding evidence versus spatial neighborhood mapping, your target's abundance, and whether your approach is antibody-driven or RNA-centric. Terminal nodes recommend specific methodologies and cross-reference the corresponding comparison tables in this review for detailed advantages, limitations, and cost considerations. Created with BioRender.com.
Figure 7. Decision guide for selecting an RNA-binding protein (RBP) identification method. To use this chart, start at the far left by defining your primary desired output: mapping the global interactome, focusing on a specific RNA target, or investigating a specific protein. Follow the branches based on your subsequent experimental constraints—such as whether you require direct binding evidence versus spatial neighborhood mapping, your target's abundance, and whether your approach is antibody-driven or RNA-centric. Terminal nodes recommend specific methodologies and cross-reference the corresponding comparison tables in this review for detailed advantages, limitations, and cost considerations. Created with BioRender.com.
Preprints 223642 g007
Table 2. In vivo methods - Cross-linking-based.
Table 2. In vivo methods - Cross-linking-based.
Method Description Advantages Limitations Cost* Time
(days)		 Throughput** Citation
RAP-MS Uses UV crosslinking and long antisense probes (90 nt) to pull down target RNA (e.g., Tau pre-mRNA) and bound RBPs; purified complexes are identified by MS. Yields high-confidence interactions due to enhanced denaturing/stringent purification; versatile for any RNA. Is expensive due to individually biotinylated probes, risks losing weaker interactions; requires sufficient material, needs high input material $$$ 5-7 Moderate (Xing et al., 2022)
iDRiP Employs cross-linking, biotinylated antisense probes, and hyper-stringent washing to eliminate weak associations and contaminants during RNA affinity purification. Offers high specificity and low background; yields more specific proteins and fewer mitochondrial contaminants. Risks losing real interactors with weaker, but functional, binding affinities due to extreme stringency. $$$ 5-7 Moderate (Chu et al., 2021)
LEAP-RBP UV cross-links proteins to RNA in living cells; isolates complexes using a lithium supplemented solvent that selectively precipitates protein-RNA complexes for MS. Cost-effective and highly efficient with a quick turnaround time. Identifies generic RNA Binding proteins, not focused on identifying proteins bound to a specific target RNA. $ 2-3 High (Kristofich & Nicchitta, 2024)
MORPH-MS Captures UV cross-linked target RNA using non-biotinylated tiling antisense oligos pulled down via hybridization to a single biotinylated "Universal oligo" and streptavidin beads. Serves as a cost-effective adaptation of ChIRP; requires only 1 expensive biotinylated oligo. (the Universal oligo) Requires optimization of the common binding sequence and antisense oligos tiling array. $ 3-4 Low-moderate (Pant & Kumarswamy, 2024)
ChIRP-MS Cross-links cells with stronger glutaraldehyde; purifies via biotinylated probes; separates proteins by SDS-PAGE, fractionates the gel, and identifies via MS. Provides a highly specific method tailored for lncRNAs. Requires a larger amount of starting input material $$$ 4-5 Low (Kim & Ma, 2021)
Biochemical Pulldown Cross-links cells with formaldehyde, hybridizes with biotinylated DNA probes, extracts with streptavidin beads, reverses crosslinking, and analyzes by MS. Captures complexes under native/in vivo conditions; maximizes protein yield for MS; offers a fast, robust protocol. Suffers from a higher amount of non-specific binding and background noise. $$ 3-4 Low (Savulescu et al., 2020)
TREX UV cross-links complexes, separating them via TRIZOL/chloroform; anneals tiling DNA oligos, degrades target RNA via RNase H, and extracts free RBPs for quantitative MS. Unbiased and cost-effective; requires no genetic manipulations and enables endogenous interactome profiling. Relies on UV crosslinking, risks potential non-specific cleavage, and requires careful oligonucleotide design. $$ 3-4 Moderate (Dodel et al., 2024)
SHIFTR UV cross-links and extracts RNA-protein complexes from interphase using acid guanidinium thiocyanate–phenol–chloroform (AGPC) organic phase separation.; uses DNA probes to pull RNA and RNase H to digest target RNA, shifting proteins to the organic phase for tandem mass tagging (TMT)-based quantitative MS. Works well with low input material; offers a superior signal-to-noise ratio; detects interactions at specific RNA regions without genetic manipulation. Remains limited by RNA abundance; may fail to detect transient or indirect interactors. $$ 4-5 Moderate (Aydin et al., 2024)
* Cost Scale: Detailed as low ($), moderate ($$), and high ($$$) based on the requirement for specialized reagents, such as individually biotinylated probes versus universal oligos. * Time & Throughput: Estimated turnaround times include cell preparation, cross-linking, purification, and mass spectrometry preparation phases.
Table 3. In vivo methods - Proximity-labeling based.
Table 3. In vivo methods - Proximity-labeling based.
Method Description Advantages Limitations Cost* Time Throughput** Citation
Proximity-CLIP Labels RNA (with 4SU) for enhanced cross-linking efficiency, utilizes a localization element (LE)-APEX2 fusion protein to label the RNA interactome; uses UV light to cross-links protein-RNA complexes. Can profile proteome and transcriptome, enables compartment-specific analysis, can capture short-lived RNA, offers high specificity, can identify precise binding sites. potential for false positives; risks cellular toxicity from hydrogen peroxide $$$ 3-5 days Moderate (Anastasakis et al., 2020)
RPL Uses dCas13b targeted by sequence-specific gRNA; fuses to engineered ascorbate peroxidase APEX2 to biotinylate electron-rich amino acids. Requires no cross-linking, sonication, or RNA engineering; works efficiently with low number of cells Risks oxidative damage, generates large complex size that may lead to reduced specificity, comparatively low efficacy. $$ 2-3 days Moderate (Lin et al., 2021)
CARPID Employs dCasRx directed by sequence-specific gRNA; fused to engineered biotin ligase (BASU) to biotinylate proximal proteins that are then pulled down using streptavidin beads Eliminates the need for cross-linking, pre-labeling target RNA, MS2 insertion, or antisense probe design; preserves the physiology of transfected cells; detects weak or transient interactions; achieves potentially broad RNA coverage using multiple gRNAs. May identify merely proximate proteins (false positives), can sterically hinder normal RBP binding or alter RNA localization, may create artifacts, long labeling time prevents analysis of RNA-protein interactions occurring over a short period of time. $$ 3-5 days Moderate (Yi et al., 2020)
CBRPP Uses dCas13 and a custom crRNA to deliver a PBL enzyme (e.g., dPspCas13b-BioID2-NES) to target RNA, which then biotinylates surrounding proteins; incorporates BioID2 (a promiscuous biotin ligase) and an NES (Nuclear Export Signal) for cytoplasmic export of the fusion protein No cross-linking required, no need to pre-label target RNA, MS2 insertion, or designing of antisense probes; maintains natural structure of target RNA and avoids RNA degradation; captures weak and transient RNA-protein interactions. May generate False positives from mere proximity, fusion protein may interfere with the binding of endogenous RBPs, long labeling time precludes capturing of short-lived RNA-protein interactions. $$ 3-5 days Moderate (Li et al., 2021)
CRUIS Uses dLwaCas13a to target specific RNA sequences; employs the proximity-labeling enzyme PafA to label surrounding RBPs by ligating PupE to lysine. No need for cross-linking, no manipulation of RNA and reduced RNA degradation, captures weak or transient interactions. May generate false positives, risks fusion proteins interfering with endogenous RBP binding $$ 2-3 days Moderate-high (Zhang et al., 2020)
MCP-APEX2 Conjugates the bacteriophage MS2 RNA stem loop; utilizes an MS2 coat protein-fused APEX2 (MCP-APEX2) to bind the stem loop while APEX2 promiscuously labels proximal proteins. No cross-linking, short labeling time to enable analysis of dynamic RNA interactomes and reduce the toxicity of hydrogen peroxide addition to cells, more sensitive than dCas13-APEX2, detects weak or transient interactions. Requires RNA modification, steric bulk of MS2 tags could block certain interactions, may result in false positives. $$ 1-2 days High (Han et al., 2020)
dCas13-APEX2 Employs a dCas13d-dsRBD-APEX2 fusion protein where dCas13d (with gRNA) targets the RNA; dsRBD stabilizes the gRNA-target RNA duplex; APEX2 proximity-labels the RNA interactome. Uses short labeling time to analyze dynamic RNA interactomes to avoid toxicity of H2O2; can detect weak or transient interactions. No crosslinking and tagging of target RNA is not required, provides higher protein specificity than MCP-APEX2, May result in false positives due to dCas13d-dsRBD fusion blocking certain interactions. $$ 1-2 days High (Han et al., 2020)
BioRBP Fuses BirA (a biotin ligase) to the phage PP7 coat protein (PP7cp); targets and binds PP7 RNA stem-loop motifs. Requires no cross-linking; detects weak or transient interactions; performs well for low-expression or unstable RNAs. May generate false positives and overexpression artifacts; requires genetic modification; risks interfering with native interactions. $$ 2-3 days High (Aeby et al., 2020)
Footnote: * Cost Scale: Detailed as low ($), moderate ($$), and high ($$$) based on the requirement for specialized reagents, such as beads, fusion protein constructs etc. * Time & Throughput: Estimated turnaround times include cell preparation, cross-linking, purification, and mass spectrometry preparation phases.
Table 4. Other experimental methods to identify RNA-binding proteins.
Table 4. Other experimental methods to identify RNA-binding proteins.
Method Description Advantages Limitations Cost Time Throughput Citation
HyPro-MS (in situ) Hybridization of digoxigenin-labeled antisense probes to target RNA and binding of HyPro, a custom-designed enzyme (fusion of engineered APEX2 and the DIG10.3 domain) that binds digoxigenin and biotinylates proximal proteins and RNA. No cross-linking required, genetic manipulation, potentially reduced artifacts caused by mislocalization and/or cytotoxicity Large labeling radius (more false positives), need to control for probe hybridization specificity, probes may compete with RBPs interacting with overlapping target sequences. $$ 2-3 days Moderate-High (Yap et al., 2022)
O-MAP (in situ) Annealing of RNA-FISH-like probes containing a universal sequence to the target RNA, and subsequent hybridization of a secondary sequence-HRP conjugate that proximity-labels surrounding molecules. Precise RNA-targeting, no need for genetic manipulation, multiomic application, low cell requirement, uses off-the-self parts. Challenging for complex FISH targets, formaldehyde fixation may disrupt the subcellular structure and can only identify molecules within the HRP labeling radius. $$ 2 days High (Tsue et al., 2024)
TPP (in vitro/ex vivo) Subjects aliquots of RNA- and vehicle-containing lysates to a temperature gradient to denature the proteins. The respective reactions are combined, and the soluble protein fractions are then analyzed through LC-MS/MS. No need for RNA modification, analysis of a wide range of interaction affinities, can detect weak/transient interactions, allow for system-level analysis (use of unpurified biological samples). Less abundant proteins may be missed due to the complexity of the lysate, provides protein-based readout only (compared to SPROX), may miss RBP hits where the RNA ligand denatures before the protein’s thermal melting temperature, may induce RNA unfolding, lower proteomic coverage (compared to SPROX). $$$ 3-4 days High (Bailey et al., 2024)
SPROX (in vitro/ex vivo) Subjects aliquots of RNA- and vehicle-containing lysates to a urea gradient to denature the proteins and a methionine oxidation reaction. The oxidation reaction is quenched, the respective samples are combined, and prepared to be analyzed through LC-MS/MS. No need for RNA modification, analysis of a wide range of interaction affinities, can detect weak/transient interactions, allow for system-level analysis (use of unpurified biological samples), provides a peptide-based readout (provides domain-specific structural information), high proteomic coverage (compared to TPP), less likely to alter the structure of the RNA molecule. Less abundant proteins may be missed due to the complexity of the lysate, depends on the protein’s methionine content. $$$ 3-4 days High (Bailey et al., 2024)
Table 5. In silico methods to identify RNA-binding proteins.
Table 5. In silico methods to identify RNA-binding proteins.
Function Method Advantages Limitations Citation
Motif Discovery and Scanning EDCNN Achieves a high average AUC score; effectively scans whole-genome datasets for structural and sequence motifs. High computational complexity and time demands (Wang et al., 2022)
RNAelem Mitigates false matches via optimized weight models; uses theoretical secondary structure space to discover novel motifs; high Area Under the Receiver Operating Characteristic curve (AUROC) . Susceptible to some false matching; profile context-free grammar (CFG) models do not account for deletions. (Miyake et al., 2024)
General Interaction Binding and Network Prediction BMF Identifies diverse sequence types (degenerate, short, complex, repetitive); models in vivo binding from in vitro data; handles synthetic and real datasets. Ignores secondary and tertiary structures; requires a core size of 3 for optimal specificity. (Sohrabi-Jahromi & Söding, 2021)
fRNC Highly customizable parameters (stringency, network size, scoring); performs well with small networks; elucidates novel mechanistic interactions. Highly dependent on experimentally derived data; requires a specific network size (~30) for optimal results. (Jiang et al., 2023)
GGCRB Fuses GCN and GAT for enhanced accuracy; integrates sequence and secondary structure features; models long-range and temporal relationships. Heavily reliant on structural data; misses dynamic/tertiary structures; poor generalizability for rare interactions. (Tang et al., 2025)
MFNN High prediction accuracy compared to traditional deep kernel models by combining matrix factorization with neural network. Struggles to accurately predict interactions for circRNAs or RBPs with sparse interaction data. (Wang & Lei, 2020)
HDRNet Robustly predicts RBP interactions across diverse cellular conditions and tissues. Encoding method restricts resolution to larger blocks rather than single nucleotides. (Zhu et al., 2023)
scRAPID Identifies multiple interaction types (RNA-RBP, RBP-RBP) as well as hub RBPs. Requires massive datasets; gene regulatory network (GRN) performance varies heavily by dataset. (Fiorentino et al., 2024)
RMDNet Features adaptable feature integration via the improved dung beetle optimization (IDBO) algorithm. Subject to inherent randomness stemming from population intelligence algorithms. (J. Zhang et al., 2025)
DeepRiPe Successfully characterizes in vivo RBP targets. Extracted filters may lack specificity for individual RBPs. (Ghanbari & Ohler, 2020)
Hierarchical
DBP/RBP 		Capable of predicting novel DBPs and RBPs simultaneously. When RBPs are included in the dataset, the model's accuracy in predicting single-stranded DNA-binding proteins (SSBs) drops significantly. (Wu & Guo, 2024)
PrismNet Accurately predicts dynamic RNA-binding protein (RBP) interactions by integrating experimental in vivo RNA secondary structure data. Validation and training are restricted to a limited subset of seven cell lines. (Sun et al., 2021)
Alternative Splicing Event Identification AGML Uses multi-view similarity matrices to preserve intrinsic data structures and identify candidate RBPs associated with alternative splicing events. Prone to computational errors when using large projection sizes. (Qiu et al., 2024)
RAIMC Reduces prediction noise using an inductive matrix completion framework and sparse similarity modeling in order to predict RBPs and alternative splicing event associations. The integration of multiple kernels can negatively impact specific prediction performance. (Qiu et al., 2021b)
WDFSMF Adaptable to other biological entities; weighting system integrates multiple data sources while minimizing background noise to help predict RBP-alternative splicing event associations. Susceptible to false positives; highly sensitive to specific parameter settings (e.g., k1 value). (Qiu et al., 2021a)
Circular RNA Specific Interaction and Feature Extraction MSTCRB Extracts multi-scale features; integrates a transformer architecture; applicable to both circRNA and linear RNA datasets. High computational complexity; pipeline dependency (Zhou et al., 2024)
DMSK Multi-view feature extraction improves recognition accuracy while reducing dimensionality in order to identify interaction sites between circRNAs and RBPs. Computationally intensive; limited by sparse circRNA data; feature reduction risks losing discriminative information. (Li et al., 2022)
circTCA Enhances feature utilization via a cross multi-head attention mechanism to predict circRNA-RBP binding sites Performance degrades (or produces errors) if an excessive number of attention heads is used. (Guo et al., 2025)
MGFCRSites Uniquely incorporates chemical molecular structural patterns for highly precise predictions of RBP binding sites on circRNAs. Susceptible to overfitting due to the reliance on residual blocks. (Liu & Zhang, 2025)
CRIECNN Utilizes four unique encoding techniques for robust feature extraction for the precise forecasting of circRNA-RBP binding sites Training of this model is constrained due to the limited amount of experimentally validated data available for circRNAs. (Lasantha et al., 2024)
RBPsuite2.0 An accessible web-based interface tool design to predict RBP binding sites across both linear and circular sequences for seven species (human, mouse, zebrafish, fly, worm, yeast, Arabidopsis). Limited to predicting RBPs that already have known binding targets. (Pan et al., 2025)
circRB Enhances binding site recognition by capturing the directional equivalence of binding orientations on circRNAs. Provides only marginal AUC improvements (~0.03) over baseline comparative models. (Wang & Lei, 2021)
SSCRB Computationally lightweight and resource-efficient for the prediction of interaction sites between circRNA and RBPs Relies on outdated programs for base-pairing generation; suffers from positive/negative sample imbalance. (L. Liu, Y. Wei, Q. Zhang, et al., 2024)
CRAFT Highly user-friendly tool for predicting circRNA function, including interactions with miRNAs and RBP as well as circRNA coding potential; automatically generates HTML pages with interactive tables and figures. Prone to downstream errors if the underlying CircExtractor tool makes incorrect predictions. (Dal Molin et al., 2022)
Pretraining and Representation Training mRNABERT Benefits from robust pretraining on one of the largest available mRNA datasets. Lack of explicit structural modeling and limitations in handling complex genomic features. (Xiong et al., 2025)
Secondary Structure Generation CRBPSA Novel architecture captures complex base-pairing and spatial positional dependencies for accurate secondary structure generation. Computational intensity of structural calculations; lack of 3D spatial context. (C. Cao et al., 2024)
Interpretability and Feature Analysis CR-deal Highly interpretable; tracks integrated gradients to explain feature binding mechanisms. Sensitive to hyperparameter tuning (e.g., learning rates and batch sizes). (Wei et al., 2025)
Ortholog Analysis PhyloPGM Reduces false positive rates by incorporating human ortholog sequences. Restricted exclusively to human species data. (Ahsan et al., 2022)
RNA Localization Prediction DeepLocRNA Blend primary sequence information with RBP binding priors in order to predict RNA localization.
Capabilities in cross-species predictions in both mouse and human.		 Optimal accuracy is currently limited strictly to miRNAs and mRNAs. (Wang et al., 2024)
Footnote: The methods from 2020-2025 publication date in English language were searched using key terms “RBP” (in title/abstract) and “In silico” (in title/abstract) in PUBMED. (("2020"[Date - Publication]: "2025"[Date - Publication]) AND (English [Language]) AND (RBP[Title/Abstract]) AND (in silico [Title/Abstract])).
Table 6. Databases for the functional annotation and structural analysis of RBPs and RNA targets.
Table 6. Databases for the functional annotation and structural analysis of RBPs and RNA targets.
Database Description Link Citations
RASP 2.0 An expanded database of 438 RNA structure probing datasets. Features tools for missing structure score imputation, RBP binding prediction, and RNA secondary/tertiary structure modeling. http://rasp2.zhanglab.net/ (Mu et al., 2025)
QUADRatlas
v1.2.2401. 		A database specifically designed to include RNA G-quadruplexes (RG4s) and any RBPs that interact with them. https://rg4db.cibio.unitn.it/ (Bourdon et al., 2023)
RMVar 2.0 A database that catalogues RNA modifications and annotations for RBP, RNA, and circular RNA interactions. Information on splicing events is also included https://rmvar.renlab.cn/ (Huang et al., 2025)
EuRBPDB 1.2 A database including information and functional annotations on eukaryotic RBPs. http://eurbpdb.gzsys.org.cn/ (Liao et al., 2020)
RBPTD
0.1.0 (2018.9) 		A database that specializes in providing information on cancer-related RBPs. http://www.rbptd.com/#/ (Li et al., 2020)
RBP-Tar
2024 		A database that organizes experimental data about RBPs from eCLIP. https://rbp-tar.biodata.ceitec.cz/ (Gresova et al., 2023)
OncoSplicing 3.0 Compiles information on RBPs that play a role in regulation of an individual alternative splicing event. http://www.oncosplicing.com/ (Y. Zhang et al., 2025)
POSTAR3 A platform capable of generating a comprehensive map of RBP binding sites using large scale CLIP-seq datasets and external functional genomic annotations. http://postar.ncrnalab.org (Zhao et al., 2022)
RBP image Database
2023 		Database mapping 301 RBPs in HepG2 and HeLa cells, featuring immunofluorescence images of RBP-organelle marker pairs alongside systematic localization annotations. https://rnabiology.ircm.qc.ca/RBPImage/ (Benoit Bouvrette et al., 2023)
NPInter v5.0 A ncRNA database featuring a RBP module to display RBP-RNA interactions. Other updates include RNA-DNA interactions and integrated SARS-CoV-2-RNA interactions. http://bidata.ibp.ac.cn/npinter5/ (Zheng et al., 2023)
RBP2GO 2.0 A database that includes information obtained from 53 studies and 105 datasets on RBPs from 13 different species. https://rbp2gov2.dkfz.de/ (Caudron-Herger et al., 2021)
SliceIt
2020 		Online database and visualization tool used to design CRISPR/Cas9 screens. It helps to map and plan the editing of protein-RNA interaction sites across the human genome. http://sliceit.soic.iupui.edu/ (Vemuri et al., 2020)
oRNAment
2020 		A compilation of putative RBP sites derived from RNAcompete and RNA Bind- n- Seq. https://rnabiology.ircm.qc.ca/oRNAment (Benoit Bouvrette et al., 2020)
EVPsort
2024 		A specialized database which focuses on ncRNAs found in extracellular vesicles and particles. Information on interactions between ncRNAs and RBPs is included. https://bioinfo.vanderbilt.edu/evpsort/ (Chen et al., 2024)
Footnote: The Databases from 2020-2025 publication date in English language were searched using key terms “RBP” (in title/abstract) and “Database” (in title/abstract) in PUBMED. (("2020"[Date - Publication]: "2025"[Date - Publication]) AND (English [Language]) AND (RBP[Title/Abstract]) AND (Database [Title/Abstract])).
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.