Preprints.org logo
Preprint
Article

This version is not peer-reviewed.

Machine Learning Analysis of Ligand Datasets for GPCR Receptors Involved in Immune Cell Trafficking and Inflammation

Submitted:

17 September 2026

Posted:

18 September 2026

You are already at the latest version

Abstract
Immune cell migration from lymphoid organs to inflamed tissues is regulated by cell-surface receptors sensitive to environmental changes. G protein-coupled receptors involved in immune cell trafficking have been used in the design of anti-inflammatory drugs providing large datasets of active compounds deposited in e.g. ChEMBL. Here, we evaluated these datasets in terms of their applicability to machine learning (ML) aimed at compound activity prediction. Mixed-receptor datasets selected based on sequence homology were used in training of deep neural networks (DNNs) and gradient boosting machines (GBMs) to develop accurate ML classifiers for receptors with sparse ligand datasets in ChEMBL such as CCR6, CCR7, CCR8, CCR9, CCR10, CXCR5, and CCRL2. This ligand-based approach to drug design was compared with structure-based virtual screening (SBVS) using refined, ready-for-VS, structural models of immune cell trafficking receptors recently implemented in GPCRVS. The strengths and weaknesses of DNNs, GBMs, and SBVS were described in terms of applicability of various data schemes. As a result, a comprehensive, multi-level approach based on publicly accessible repositories has been proposed to facilitate immunomodulatory drug design.
Keywords: 
artificial intelligence
;  
machine learning
;  
deep learning
;  
gradient-boosting machine
;  
G protein-coupled receptors
;  
drug discovery
;  
chemokine receptors
;  
atypical chemokine receptors
;  
immune system
;  
immune-mediated inflammatory disease
Subject: 
Biology and Life Sciences  -   Immunology and Microbiology

1. Introduction

Immune cell trafficking is strictly regulated and highly organized to provide an immediate and resilient response to pathogens [1]. However, altered trafficking of immune cells and pathogenic immune cell circuits leads to development of chronic inflammatory conditions [2]. Modulation of T cell circulation across tissues has recently been proposed as a new therapeutic strategy for immune-mediated inflammatory diseases (IMID) [3,4]. Chemokines and their concentration gradient are key regulators of immune cell motility in complex tissue environments [5].
Recently, we developed GPCRVS (https://gpcrvs.chem.uw.edu.pl), a web service based on machine learning (ML) for virtual screening against chemokine receptor drug targets regulating the immune cell trafficking [6]. In addition to CC, CXC, and atypical chemokine receptors (ACKRs), receptors for chemerin, the fractalkine receptor CX3CR1, and complement system receptors C3aR and C5aR1/2 were also included in GPCRVS to enable the multiscale drug screening for the development of novel anti-inflammatory and immunomodulatory therapies.
Chemokine receptors are among the most extensively studied inflammatory G protein-coupled receptors (GPCRs) and signal through chemokine ligands to regulate leukocyte trafficking [7]. Chemokine receptors fall into two main subfamilies: atypical receptors that do not bind G proteins (ACKRs) and conventional receptors (cCKRs) [8]. The latter are further divided into subtypes based on the arrangement of conserved cysteine residues in their respective chemokine ligands: CC receptors (CCRs), CXC receptors (CXCRs), CX3C receptors (CX3CRs), and XC receptors (XCRs) [9]. Chemokine receptors have been implicated in diseases such as atherosclerosis (CCR2, CXCR1, CXCR2), autoimmune deficiency syndrome (CCR5, CXCR4), multiple sclerosis (CCR1, CCR2, CCR5, CXCR3), rheumatoid arthritis (CCR1, CCR2, CCR5, CXCR3) [3], as well as cancers (e.g., CCR1 in breast cancer [10]; CCR2 in pancreatic cancer [11]; CXCR4 in ovarian cancer [12]).
Chemerin-like receptor 1 (CMKLR1) has been linked to both pro- and anti-inflammatory effects through its activation via two distinct endogenous ligands: chemerin and resolvin E1 [13]. Chemerin is a chemoattractant involved in regulating innate immunity and inflammation. Signaling through this axis leads to a proinflammatory effect involving the trafficking of dendritic cells and macrophages, while also contributing to adipogenesis and angiogenesis. On the other hand, interactions with resolvin E1 have been shown to attenuate leukocyte infiltration and suppress proinflammatory gene expression [13]. Among others, CMKLR1 has been implicated in multiple sclerosis [14] and clear cell renal cell carcinoma [15]. A closely related receptor, CMKLR2 (formerly GPR1), similarly binds chemerin but is less well studied [16].
In addition to CMKLR1 and CMKLR2, chemerin also binds to C-C chemokine receptor-like 2 (CCRL2) [17]. Though it was originally thought to bind certain chemokines as well, these observations remain unconfirmed. Like the receptors described above, CCRL2 plays a role in leukocyte trafficking; however, unlike conventional receptors, chemerin binding to CCRL2 does not lead to signaling or internalization. Instead, CCRL2 acts as a chemerin-capturing receptor, increasing its local concentration and facilitating chemerin–CMKLR1 interactions by presenting the peptide’s unbound C-terminus. CCRL2 also forms dimers with CXCR2 and regulates neutrophil migration [17].
Complement component 3a (C3aR) and 5a (C5aR1 and C5aR2) receptors are primarily known for their roles in complement-mediated immunity. The endogenous ligands of these receptors, the anaphylatoxins C3a and C5a, are generated during complement activation, which can be triggered, e.g., by microbial infiltration or tissue injury [18,19]. C3aR activation by C3a regulates the inflammatory response by, e.g, inducing histamine secretion, smooth muscle contraction, and the chemoattraction of leukocytes [20]. C5aR1 similarly mediates proinflammatory signals, contributing to immune cell chemotaxis and cytokine release [21]. Conversely, C5aR2 has been shown to play a dual pro- and anti-inflammatory role in different disease models [22,23]. Dysregulation of the complement system contributes to neuroinflammation, autoimmune diseases, chronic inflammatory and metabolic diseases, cardiovascular diseases, and cancer [24].
Despite the structural similarity shared across class A GPCRs, chemokine and related inflammatory GPCRs differ in their physiological roles and signaling outcomes. As such, to develop selective drugs, receptors must be analyzed individually, and subtle structural differences must be considered. Achieving high selectivity and specificity toward a specific pathway is essential to modulating only pathogenic signaling without affecting protective immune functions [25,26]. This issue can be addressed by developing AI-based virtual screening tools that are capable of detecting subtle structural differences in both receptors and ligands.
Numerous attempts have been made to facilitate GPCR drug discovery due to their profound impact on the regulation of physiological networks and high druggability. Structure-based approaches like molecular docking, e.g. AutoDock Vina [27] or Schrödinger Glide [28], have been significantly enriched with ligand-based approaches involving machine learning. ML-based approaches include GPCRVS [6], SwissTargetPrediction [29], pdCSM-GPCR [30], EnGCI [31], and DeepGPCR [32]. Among these, SwissTargetPrediction is perhaps the best-known online tool for predicting compound affinity to over 3000 targets across humans, mice, and rats. SwissTargetPrediction, similar to GPCRVS, is based on data deposited in ChEMBL [29]. pdCSM-GPCR uses graph-based signatures, or feature representations that capture a graph's structure and properties by summarizing patterns of nodes, edges, and their attributes into a compact form for machine-learning models. For molecules, covalent bonds are represented as edges and atoms as nodes. pdCSM-GPCR predicts compound activity for 36 GPCRs of classes A, B, C, and F [30]. EnGCI is a framework that predicts receptor-compound interactions by integrating graph isomorphism networks for GPCR feature extraction and convolutional neural networks for compounds [31]. DeepGPCR, like pdCSM-GPCR, is a graph-based tool for predicting compound affinity. DeepGPCR has been used to identify active compounds for GPR35, an orphan G protein-coupled receptor, and GLP1R, the glucagon-like peptide-1 receptor [32]. Beyond GPCR-specific methods, general-purpose binding-site predictors such as MPBind [33] have also shown the value of large, pooled datasets. However, this approach identifies binding-site residues broadly and does not address the activation-state-specific ligand selectivity that GPCRVS focuses on.
Here, we present an updated version of GPCRVS focused on receptors involved in immune cell trafficking, which represent important drug targets for the pharmacotherapy of immune-mediated inflammatory diseases. GPCRVS, first released in 2023, has been updated with newly retrieved ChEMBL datasets and expanded to include additional receptors (e.g., ACKRs, chemerin receptors, and complement system receptors), with improvements to both predictive models and receptor structures, resulting in enhanced accuracy and broader applicability in drug design for IMIDs. Notably, among many GPCRs that are involved in inflammatory conditions, chemokine receptors (Table 1) represent drug targets that trigger stable polarization and long-range chemotaxis of immune cells while others act differently, e.g., sphingolipid S1P acting through S1PR1 triggers a transient polarization only and no significant displacement contributing to a brief transmigration step to exit into efferent lymphatics according to [34].

2. Results

Despite using the same training datasets, the performance of DNNs and GBMs varied (Table 2, Figure 1 and Figure 2, Supplementary Table S1, Figure S1), suggesting that these two algorithms represent intrinsically different but complementary approaches to machine learning. DNNs are commonly used for classification tasks and handling categorical variables, whereas regression tasks requiring precise qualitative prediction are often performed with GBMs [114]. In our results, we also observed these trends, showing that GBMs slightly outperform DNNs in the case of unambiguously labeled, structured data (high-quality but often limited), e.g., in the case of CCR2-active, CCR3-active, CCR9-inactive, CXCR3-inactive, CXCR4-inactive and active (Table 2, Supplementary Table S2). However, DNNs often win in more difficult cases with ambiguous or redundant data labeling, often including ‘me-too’ drugs or drug-like compounds such as CCR1-inactive, CCR5-inactive, CCR6-inactive (Table 2, Supplementary Table S2). In the CX3CR1-inactive case with 408 compounds out of 418 with the assigned pChEMBL value of 6.5 (an example of a ‘me-too’ compound dataset) DNNs also slightly outperformed GBMs. Both DNNs and GBMs performed exceptionally well, e.g., in the case of the C5aR-inactive and CXCR5-inactive datasets (RMSE 0.160 and 0.489, accuracy GBMs - 85.7 % and 75 %, DNNs - 86.7 % and 81.3 %, respectively), which included substantially different compounds with varied activity (Table 2, Supplementary Table S2). Although a weak correlation was observed between the number of training-set compounds and test-set accuracy (Supplementary Figure S5), variability in classifier performance may also stem from factors such as assay heterogeneity.
Exact datasets with compounds significantly varied in both structure and activity, CCR3-inactive, CCR3-active, CCR4-active, CCR9-inactive, CXCR3-inactive (Supplementary Table S2) were slightly better exploited with GBMs than with DNNs. This included both larger datasets, such as CXCR3-inactive, and smaller, more limited datasets, such as CCR3-active or CCR4-active. Fuzzy, but rather large compound datasets, such as CCR5-inactive, CCR4-inactive, CXCR1-inactive, CCR6-inactive, CCR7-inactive, CXCR2-inactive, could still be used as training datasets for DNNs, providing a maximal prediction accuracy of 68.9 % (the respective RMSE was equal to 3.68, accuracy – 58.9 %, GBM results for CCR5-inactive). The fuzziness of datasets was due to the same pChEMBL values being assigned to many, substantially different compounds that represented diverse scaffolds, easily distinguished based on their fingerprints (e.g. CCR4-inactive, CXCR1-inactive, Supplementary Table S2). In the case of ambiguous data, e.g., CCR10-inactive and C3aR-inactive, DNNs also outperformed GBMs (Table 2). Here, ambiguity means that highly varied activity was assigned to nearly the same compounds with very similar fingerprints (Supplementary Table S2). Results of DNNs and GBMs mostly overlapped in the case of varied training datasets of average number of data points, such as CX3CR1-inactive, C5aR-active, CXCR5-inactive, CCR9-inactive, CCR5-active, also in the case of mixed receptor datasets such as CCR8-active, CXCR5-active, CCR6-active, CCR10-active, but were complementary for limited or fuzzy datasets (see above). The largest differences between performance of DNNs and GBMS for CMKLR1-active, CCR6-inactive, and CCR10-inactive are visible in Figure 1.
Only CCR8 receptor datasets included only orthosteric ligands or only allosteric modulators (Table 2, Supplementary Table S3). Consequently, results obtained for control datasets were similar for orthosteric and allosteric modes implemented in GPCRVS for receptors involved in immune cell trafficking (Figure 2, Supplementary Figure S2 and Table S1). The largest deviations from the experimental values, computed as the absolute difference between pChEMBLexperimental and pChEMBLpredicted, and exceeding 4.0, were observed for molecular docking in the case of CXCR4-inactive and CCR5-inactive. In the remaining cases, machine learning was significantly better than molecular docking, though low DNN prediction accuracy or large GBM prediction RMSE correlated with decreased molecular docking accuracy (e.g., CCR2-inactive, CXCR1-inactive, CXCR2-inactive). For most of the receptors, the pChEMBL prediction error was within the 0.0 – 4.0 range, meaning that a highly active ligand was hardly ever predicted as an inactive or weakly active ligand, or in reverse. The best ML predictions, across at least 20-compound control datasets, within the 0.0 – 2.0 range of the pChEMBL difference, were obtained for C5a, CX3CR1, CMKLR1, CCR9, CXCR3, and CXCR4 receptors.

3. Discussion

The data preparation step preceding ML algorithm training is the most time-consuming step, especially for error-prone, unstandardized, ambiguously labeled, fuzzy, or sparse experimental data from functional assays. These problems are not caused by experimental limitations, as currently drugs can be easily tested against thousands of GPCR variants with multiplexed assays of variant effect (MAVE) [115], but they are rather caused by difficulties in correct transformation of experimental observables to precise numerical variables [116]. Here, we showed that thorough curation applied to ChEMBL-retrieved datasets including GPCR ligands for receptors involved in immunity ensures their applicability in machine learning aimed at anti-inflammatory drug design. Nevertheless, the size of ChEMBL datasets for GPCRs that could be used for ML is not uniformed (Table 2), ranging from large datasets with thousands compounds to highly limited datasets with less than 100 compounds that often provide overfitted ML models. The data enrichment step using other databases such as PubChem or Enamine could be considered though there is already a certain overlap between these databases and ChEMBL. What is more, many active compound datasets are deposited in restricted-access, non-academic repositories.
The difficulty of drug design for immune-mediated inflammatory disease lies in the number of drug targets involved and their influence on both disease conditions and self-defense mechanisms that should be left intact. Closely related receptors sharing a conserved seven-transmembrane architecture (Figure 3) can still differ in ligand-binding residues, active ligands, and downstream signaling. One way to distinguish signaling pathways that should be modulated in treatment from those that should be left intact is to design highly selective drugs [25] that also demonstrate high specificity and a certain signaling bias toward a beta-arrestin or G protein subtype pathway [26]. For this reason, the development of AI-based tools that integrate data from multiple sources and are more sensitive to subtle differences between receptor subtypes or signaling biases is of a high priority.

4. Materials and Methods

4.1. Ligand Datasets for LBVS

To prepare ligand datasets for ML-assisted, ligand-based virtual screening (LBVS) active ligands for GPCR receptors involved in immune cell trafficking and inflammation were retrieved from functional assays from ChEMBL [117] (accession date – not later than July 1, 2025). Data point distributions for these exact and fuzzy compound datasets (see 2. Results) was shown in Supplementary Table S2. RDKit and Python scripts were used for data curation in terms of: duplicates with the same SMILES, missing SMILES data, missing pChEMBL, data ambiguity, C-terminal modifications of peptide ligands as described in [6], and data annotation involving selective/unselective ligand split. Exact compound duplicates with the same SMILES or ChEMBL ID but with the varied activity assignment obtained from different assays were removed, leaving the data points with the highest pChEMBL values. The ligand type split was based on the text search with ‘agonist’, ‘inverse agonist’, ‘antagonist’, ‘allosteric modulator’ terms. The active/inactive threshold for the compound activity represented as pChEMBL values were assigned as follows: inactive – pChEMBL below 4.0, weakly active – pChEMBL from 4.0 to 6.0, moderately active – pChEMBL from 6.0 to 8.0, and highly active – pChEMBL above 8.0.
The CCR8 allosteric modulator datasets included compounds with pChEMBL in the range of 4.4 to 6.5, while the range of CCR8 orthosteric datasets was 4.58 to 9.52, because allosteric modulators tend to bind to their targets with a lower potency than orthosteric ligands [118]. All the remaining datasets for other receptors included unassigned ligands in terms of allostery/orthostery. In such cases, one-receptor ML classifiers for both allosteric and orthosteric modes were trained using the same ligand datasets (Supplementary Table S3).
Notably, ChEMBL datasets for GPCRs partially overlap due to common lack of target selectivity for many GPCR ligands. In addition, mixed-receptor datasets (see below) included repeated compounds from three receptor datasets. As a result of these limitations of currently available ChEMBL datasets, training datasets included on average 75 % of selective compounds and 25 % of unselective compounds included in more than one receptor dataset. As for the overlap between inactive receptor and active receptor datasets (inhibitors vs. agonists), on average 6.7 % of ligands from inhibitor datasets were also included in active receptor datasets. This was mainly caused by the lack of the ligand type assignment in the assay data retrieved from ChEMBL. This certain overlap in ChEMBL datasets for different receptors or for different receptor activation state could be perhaps removed, but not if obtained compound datasets were too small for ML like here. Currently, many of ChEMBL datasets with ligands of immune system GPCR receptors are too limited to perform such an additional task in the data curation step.
The typical 80/10/10 random split was applied to ligand datasets providing training, validation, and test datasets (Pandas Dataframe function sample()). In the case of limited compound datasets for certain receptors, only the so-called control datasets (Figure 2) were used to assess the prediction accuracy of the ML models obtained for these receptors. These control datasets included either test datasets representing 10 % of the curated ChEMBL datasets from the 80/10/10 random split (see above) or a sparse dataset if only a few active compounds were available for a certain receptor in ChEMBL. In the case of receptors for which compound datasets were too sparse to be used in training, mixed-receptor datasets were used. These mixed-receptor datasets included compound datasets for 3 receptors with the highest sequence identity to the receptor with that sparse ChEMBL dataset according to ClustalOmega. These mixed-receptor datasets were used for training ML models for CCR6-active (CXCR3, CCR4, CX3CR1), CCR7 – both (CCR2, CCR4, CCR5), CCR8-active (CCR1, CCR4, CX3CR1), CCR9-active (CCR1, CCR4, CCR5), CCR10-active (CCR3, CXCR3, CX3CR1), CXCR5-active (CCR4, CXCR3, CX3CR1), and CCRL2-active (CCR2-active, CCR1-active, CCR5-active). For ACKR5 (UniProt ID: O15218, ChEMBL ID: CHEMBL4356), both active and inactive, the CXCR7 (ACKR3) ML model was used in the activity prediction, because it was the only atypical chemokine receptor other than ACKR5 used in this study. According to [119] another possibly atypical receptor CCRL2 (UniProt ID: UniProt ID: O00421, ChEMBL ID: CHEMBL2321627), still awaits validation and the appropriate taxonomy assignment. Notably, there is a certain discrepancy in the taxonomy of atypical chemokine receptors requiring systematic annotation [119,120]. ACKR4 was not included in the study because of the lack of ChEMBL compound datasets, and ACKR1 and ACKR2 datasets included only 1 and 6 compounds and therefore were also excluded. For the same reasons, for the ligand activity prediction against C5aR2, the C5aR1 ML model was used (Supplementary Table S3).
Open Babel and Maestro were used in structure-based virtual screening (SBVS) to reconstruct 3D ligand structures. No additional ligand minimization was performed if Open Babel failed to convert 2D SMILES into 3D ligand structures to ensure the automatic pipeline without the manual ligand structure refinement. Failed compounds at this step were discarded contributing to slight differences between testing and control datasets (see above).

4.2. Machine Learning and Model Training

Two diverse ML algorithms were trained using the aforementioned ligand datasets in recognition of active ligands: deep neural networks and gradient boosting machines [6]. TensorFlow/Keras and LightGBM were used for their implementation, respectively [6]. One-receptor ML models (Supplementary Table S3) trained using the above GPCR ligand datasets were implemented in GPCRVS, a web service for GPCR virtual screening [6]. 10-fold cross-validation was used in the ML model training, hyperparameter tuning, and model evaluation. Results of 10-fold cross-validation in comparison with results obtained for test datasets, not used in either training or validation were shown in Supplementary Figure S1. The performance of GBMs was assessed with the RMSE metrics at first, disproportionately penalizing large errors [121], and to compare with DNNs also with accuracy after transforming the continuous values for the compound activity into the discrete values corresponding to the activity classes defined for DNNs previously in [122]. To compare ML results with AutoDock Vina results, docking scores were also transformed into pChEMBL values (see below 4.3, Supplementary Figure S3), and later into the same 6 activity classes as for DNNs: 1 – pChEMBL above 8.0 – highly active compounds; 2 – pChEMBL 7.0 – 8.0 – medium-active compounds; 3 – pChEMBL 6.0 – 7.0 – medium-active compounds; 4 – pChEMBL 5.0 – 6.0 – low-active compounds; 5 – pChEMBL 4.0 – 5.0 – low-active compounds; 6 – pChEMBL below 4.0 – inactive compounds. In [6] a more detailed description on data curation, feature generation, and model training is included. The agreement between top feature structural scaffold detected with GBMs (the feature importance detection with the gain method in LightGBM) and indeed observed in active compounds of immune receptors was shown in Supplementary Table S4.

4.3. Structure Modeling for SBVS

In addition to curated ligand datasets described above, a complete set of structural models, both in active and inactive conformations, was prepared for all included receptors regulating immune cell trafficking (Figure 3). Receptor structures available in PDB were refined in terms of loops and N- and C-termini using either MODELLER or Rosetta loop modeling, following the procedure described in [6]. Receptor structures not available in PDB were prepared with either the homology modeling procedure [6] or using structural models obtained from AlphaFold [123], I-TASSER [124], GPCRdb [125], or Robetta [126] that were subsequently refined similarly to PDB structures. I-TASSER, GPCRdb, and Robetta models were generated for a previous study [122], in which their quality was assessed by docking decoy compounds and generating receiver operating characteristic (ROC) curves. The selected AlphaFold models were obtained directly from UniProt and assessed in PyMOL [127] to determine whether they were active or inactive by examining micro- and macroswitch conformations. Low-confidence sequence regions of the AlphaFold models (pLDDT < 50) were refined using MODELLER (loops of less than 20 residues) or Rosetta CCD loop modeling (for loops more than 20 residues long). The overall quality of the generated models was assessed with MODELLER – DOPE, zDOPE, molpdf scores, and with Autodock Vina – receiver operator curves (ROC) for the active ligand enrichment with the area under the curve (AUC) values (Figure 3, Supplementary Figure S4, Table S5). The ROC curves were generated using the pROC package [128] in R [129], with the best-mode AutoDock Vina binding affinity for each ligand as the classifier score and known actives versus decoys as the reference classification. AUC values were calculated using the roc() function.
Ready-to-use, SBVS-fitted, refined structures of chemokine receptors CCR1–CCR6 had previously been prepared for GPCRVS, version 1.0 [6]. The template structure for the CCR1 active-state model had been retrieved from GPCRdb [125] (accession date: December 1, 2022). Here, its N-terminus was extended to include a cysteine that forms a disulfide bridge with C7.25 (Ballesteros-Weinstein numbering scheme for class A GPCRs). The CCR1-inactive model had been generated by Robetta [126] (accession date: December 1, 2022). The N-terminus was extended similarly to CCR1-active; likewise, the C-terminus was extended to match the CCR1 active conformation. The CCR2-active model had been downloaded from GPCRdb (accession date: December 1, 2022). Here, the N-terminus was extended by two amino acids, and a disulfide bridge was included between the N-terminal cysteine and C7.25. The 6GPX [130] PDB structure was used as the CCR2-inactive model, as it had previously been determined to be the highest-quality structure among those available [122]. However, this structure is missing intracellular loop 2 (ICL2), ICL3, and helix 8 (H8). Here, MODELLER was used to transfer these fragments from CCR2-active, and the N-terminus was extended similarly to that of CCR2-active. CCR3-active had been downloaded from GPCRdb (accession date: December 1, 2022). Again, the N-terminus was extended to include a cysteine that forms a disulfide bridge with C7.25. CCR3-inactive had been generated by Robetta (accession date: December 1, 2022). Here, the N- and C-termini were extended to match the CCR3-active sequence and to include the aforementioned disulfide bridge. CCR4-active had been generated by I-TASSER [124] (accession date: December 1, 2022). The N-terminus was again extended to form a disulfide bridge with C7.25. CCR4-inactive had been generated by Robetta (accession date: December 1, 2022). The N- and C-termini were extended to match the CCR4-active sequence. 7O7F [131] had previously been used as the CCR5-active model; no modifications were made to the structure. 4MBS [132] had been used as the CCR5-inactive model. Here, its N-terminus was extended to match 7O7F and include the disulfide bridge with C7.25. MODELLER was similarly used to extend the C-terminus to match, and then to model the missing ICL3. 6WWZ [133] had previously been used as the CCR6-active model; no further modifications were made here. In the previous version of GPCRVS, CCR6-inactive was a homology model. Here, it was replaced with the new 9D3G structure [134]. Using 6WWZ as a template, the N-terminus was extended to match CCR6-active and include the cysteine bridge to C7.25. The missing regions of ICL2, ICL3, extracellular loop 2 (ECL2), and ECL3 were reconstructed using MODELLER.
Chemokine CC receptors CCR7–CCR10 were not included in the previous version of GPCRVS. To date, no active-state structure has been solved for CCR7, but it shares a high sequence identity with CCR6. For this reason, 6WWZ was used as the template for CCR7-active; additionally, ECL2 and H8 were taken from the inactive-state CCR7 structure 6QZH [135] to make the CCR7 models uniform. 6QZH was also used as the template for CCR7-inactive, with part of transmembrane helix 1 (TM1) and the N-terminus being transferred from 6WWZ. The missing fragments of ICL2, ICL3, and ECL3 were reconstructed and the C-terminus was extended using MODELLER. The structure of CCR8-active was based on 8KFX [136], as it had the best resolution of the available active-state structures of CCR8. The extracellular regions of TM1 and the N-terminus were extended using 6WWZ as a template. The AlphaFold model retrieved from UniProt was used as a template for CCR8-inactive. Homology models of CCR9-active were generated using 6WWZ as the template, with H8 transferred from the inactive-state CCR9 structure 5LWE [137]. CCR9-inactive was reconstructed using 5LWE as a template, with ECL2 transferred from 6WWZ. MODELLER was additionally used to extend the N-terminus and to reconstruct ECL1. The AlphaFold [123] model retrieved from UniProt was used as a template for CCR10-active. Due to high sequence identity with CXCR3, CXCR3-inactive (described below) was used as a template for CCR10-inactive.
CXCR1–CXCR3 were included in the previous version of GPCRVS. 8IC0 was used as a template for CXCR1-active, and the C-terminus was extended to match CXCR1-inactive. A model of the inactive-state CXCR1 was obtained from GPCRdb (accession date: December 1, 2022). The N-terminus was extended to match CXCR1-active and to enable the inclusion of a disulfide bridge to C7.25. CXCR2-active was based on 6LFO [138] and the C-terminus was extended to match CXCR2-inactive. A model of the inactive-state CXCR2 was obtained from GPCRdb (accession date: December 1, 2022). Again, the N-terminus was extended to match CXCR2-active and to enable for the inclusion a disulfide bridge to C7.25. In the first release of GPCRVS, both the inactive- and active-state models of CXCR3 were homology models. Since then, the structure of CXCR3 in both activation states has been resolved. The first CXCR3-active model was replaced with 8HNM [139], which has the best resolution of all available active-state structures and has a bound small-molecule ligand. The truncated TM1 was extended using TM1 of the 8K2W [140] inactive-state structure as a template, and the N-terminus was additionally extended to include the disulfide bridge with C7.25. The original CXCR3-inactive model was replaced with 8K2W, as it has the best resolution of the two available inactive-state structures. The truncated H8 was extended using H8 from the active-state 8HNM as a template, and again, the N-terminus was extended to include the disulfide bridge with C7.25.
CXCR4-active was based on 8U4O [141]. Though other structures with higher resolutions were present in PDB, they either included no ligand, suggesting the orthosteric binding site would be less suitable for molecular docking (8U4N [141]), or a shorter N-terminus (8U4P [141] and 8K3Z [142]), or a more irregular β-hairpin in ECL2 (8K3Z [142]). Here, the C-terminus was slightly extended to match CXCR4-inactive. CXCR4-inactive was based on 3ODU [143], as it has the best resolution of all the available structures. As 3ODU lacks ICL3, this fragment was transferred from 8U4O. 3ODU also lacks a regular H8, similarly to other inactive-state CXCR4 structures [143], but not the active-state structures of CXCR4 (e.g., 8K3Z). However, to ensure both the active- and inactive-state structures are uniform for molecular docking, H8 was rebuilt using 8U4O as a template. The N-terminus was extended to match the CXCR4-active sequence. CXCR5 lacks a solved structure for either its active or inactive state. It shares high sequence identity with CXCR3, and a homology model of CXCR5-active was generated using CXCR3-active as a template. The AlphaFold model retrieved from UniProt was used as a template for CXCR5-inactive. 7SK7 [144] was used as a template for the active state CXCR7 (also known as ACKR3) because it has the best resolution structure bound to a small-molecule ligand. MODELLER was used to reconstruct the missing ICL1, ICL2, and ICL3. The missing H8 was transferred from CXCR7-inactive. Since the inactive-state structure of CXCR7 has not yet been resolved, CXCR2-inactive was used as a template for homology modeling, due to the high sequence identity shared by these receptors. The N-terminus and ECL2 were transferred from CXCR7-active to make the models uniform. CX3CR1 was based on 7XBW [145], and the N-terminus was extended to include the disulfide bridge with C7.25. CX3CR1-inactive did not have a resolved structure in PDB, and instead, the CCR2-inactive 6GPX structure was used as a template for homology modeling with MODELLER. The missing ECL2, ICL2, and ICL3 were transferred from CX3CR1-active. The N-terminus was additionally extended to include the aforementioned disulfide bridge.
CMKLR1-active was based on 7YKD [146], which has the best resolution of the two, available in PDB structures. The N-terminus was reconstructed using 8XGM [147], a structure of chemerin bound to the active-state CMKLR2, as the template. MODELLER was used to reconstruct the missing fragment of ECL2, and a conformation in which this fragment did not cover the orthosteric binding site of the receptor was selected for the final model. The AlphaFold model retrieved from UniProt was used as CMKLR1-inactive, however the N-terminus, C-terminus and a fragment of H8, and the aforementioned ECL2 fragment were removed due to low confidence scores. The N-terminus and ECL2 fragment were then transferred from CMKLR1-active to ensure uniformity between the two models. CMKLR2-active was reconstructed using 8XGM as a template. Though 8JJP [147] has a better resolution, this structure includes only a short chemerin fragment bound to the receptor, and the receptor itself lacks an N-terminus. Because of this, some of the residues at the extracellular side of the receptor had slightly different conformations than in 8XGM. 9L3Y [148], on the other hand, has a worse resolution than the other two, lacks a fragment of the N-terminus, and certain regions, (TM4, ECL2, and H8) are disordered. As CMKLR2 does not have a resolved inactive-state structure, CMKLR1-inactive was used as a template, and the N-terminus was transferred from CMKLR2-active. CCRL2 lacks a resolved structure for both its active and inactive conformations. As such, CCR1-active was used to model CCRL2-active, and H8 was transferred from CCLR2-inactive. The AlphaFold model retrieved from UniProt was used as the CCRL2-inactive model, and the N-terminus, ECL1, ECL2, ECL3, ICL2, and ICL3 were transferred from CCRL2-active to ensure uniformity. ACKR5, similarly, lacks resolved structures, but shares high sequence identity with CXCR7. For this reason, CXCR7-active and CXCR7-inactive were used as templates for the homology modeling of ACKR5-active and ACKR5-inactive, respectively.
C3aR-active was based on 8IA8 [149] which has the best resolution of the available active-state structures. C3aR includes a long, possibly disordered fragment in ECL2, comprising 144 residues, which is not available in any of the solved structures. For structure prediction purposes, this fragment (from K45.64 to V5.31) was removed in both the active- and inactive-state models. Additionally, the N-terminus of C3aR-active was extended by eight residues. The AlphaFold model retrieved from UniProt was used as C3aR-inactive after removal of the low-confidence areas, including the N- and C-termini, and the aforementioned ECL2 fragment. C5aR1-active was based on 7Y64 [150]. Though 7Y67 [150] has a slightly better resolution, it has a shorter N-terminus, lacks part of ECL2, and has only a fragment of the C5a protein in the active site. H8 was extended to ensure uniformity with C5aR1-inactive. C5aR1-inactive was based on 6C1R [151] as it has the best resolution of all the inactive-state structures. The N-terminus was extended to match C5aR1-active, and MODELLER was used to model the missing ICL3 and ICL4. C5aR2 does not have any resolved structures but because it shares a high sequence identity with C5aR1, C5aR1-active and C5aR1-inactive were used as templates for C5aR2-active and C5aR2-inactive, respectively. For both receptor states, ICL3 and ICL4 were modeled using MODELLER, and an inverted H8 conformation based on 6C1R was generated for C5aR2-active.
The structures described above were used in rigid-receptor, flexible-ligand molecular docking with AutoDock Vina, version 1.2.4. A preparation step, including the addition of hydrogen atoms and assignment of partial charges, was performed using the previously described approach based on OpenBabel, version 3.2.0 and MGLTools 1.5.7. The molecular docking module has been implemented in GPCRVS. To compare results of AutoDock Vina with results obtained with either GBMs or DNNs, docking scores were transformed to pChEMBL values using following ranges based on histograms obtained for all GPCR ligands from control datasets (Supplementary Figure S3). Inactive compounds with pChEMBL smaller than 4.0 were assigned to docking scores smaller than -2.0 kcal/mol or -4.0 kcal/mol to allosteric and orthosteric binding modes, respectively. Weakly active compounds with pCHEMBL in the range between 4.0 and 6.0 were assigned docking scores between -5.0 and -2.0 kcal/mol, and between -6.0 and -4.0 kcal/mol, respectively. Compounds of medium activity with pCHEMBL between 6.0 and 8.0 were assigned docking scores between -7.0 and -5.0 kcal/mol, and between -8.0 and -6.0 kcal/mol, respectively. Highly active compounds of with pCHEMBL above 8.0 were assigned docking scores below -7.0 kcal/mol and below -8.0 kcal/mol, respectively. These docking score assignments to pChEMBL values were based on typical ranges that GPCR ligands demonstrate in molecular docking to orthosteric active sites (between -12.0 and -4.0 kcal/mol) and allosteric sites (between -10.0 and -2.0 kcal/mol). AutoDock Vina scoring function [152] corresponds to the compound affinity for the target, measured experimentally in binding assays. While docking scores are not directly indicative of the compound activity, measured experimentally in functional assays, it is commonly assumed that every active compound would also demonstrate high affinity for the receptor, whereas the opposite would not necessarily be true [153].

5. Conclusions

The most important advantage of ChEMBL datasets [117] is their open-accessibility, constant updates guaranteed by EMBL, and possibilities to examine each ligand dataset in detail on the ChEMBL website to analyze outliers, negative data points, and ambiguous, fuzzy, or simply inconsistent data from diverse sources. We observed that a combination of LBVS and SBVS implemented in GPCRVS, using curated ligand datasets retrieved from ChEMBL and refined, ready-for-VS, receptor structures, efficiently compensates inaccuracies in these both approaches to drug design. We showed that machine learning based on molecular fingerprints enables to detect ambiguity and fuzziness of datasets and to assess their overall quality. Our method is aimed at fast, online, rough but accurate assessment of the compound activity and selectivity against different types and subtypes of GPCR receptors involved in the immune cell trafficking used as drug targets for IMIDs.

Supplementary Materials

The following supporting information can be downloaded at: Preprints.org, Table S1. Performance comparison of three modules – DNNs, GBMs, Autodock Vina, implemented in GPCRVS, for control datasets; Table S2. Data point distributions for compound datasets of immune system GPCRs in ChEMBL obtained with GBMs; Table S3. ML classifiers for immune system GPCR drug targets implemented in GPCRVS; Table S4. Examples of compound scaffolds detected by GBMs as the top features guiding the activity prediction; Table S5. A comparison of the model quality assessment performed with MODELLER and the ROC curve analysis for structural GPCR models; Figure S1. Results of 10-fold cross-validation performed for DNNs and GBMs in the ligand activity prediction; Figure S2. Differences between experimental and predicted pChEMBL values for control dataset compounds – the allosteric mode of GPCRVS; Figure S3. Histograms of AutoDock Vina docking scores for ligands of GPCR receptors involved in immune cell trafficking; Figure S4. ROC curves for refined PDB structures and homology models of immune system GPCRs; Figure S5. Test-set accuracy vs. training set size.

Author Contributions

Conceptualization, D.L.; methodology, D.L. and P.D.; software, D.L. and P.D.; validation, D.L. and P.D.; formal analysis, D.L. and P.D.; investigation, D.L. and P.D.; resources, D.L.; data curation, D.L. and P.D.; writing—original draft preparation, D.L. and P.D.; writing—review and editing, D.L. and P.D.; visualization, D.L. and P.D.; supervision, D.L.; project administration, D.L.; funding acquisition, D.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Data Availability Statement

The original data presented in the study are openly available in Zenodo and on the GPCRVS website (see below). The datasets presented in the study are openly available in Zenodo: https://zenodo.org/records/20492145?token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjVjYmRlNTczLWZkMWUtNDllMy1iNDFmLWE5MTU2ZmZiYTFkZiIsImRhdGEiOnt9LCJyYW5kb20iOiJhMzg2OGZhODA1ZTk4OWY5MjM2NmNmZjBkNmM2ODExYiJ9.nCua9KTQTYm0pF8Iy44U2ne1fyF_X1DQVxiNBLzBttTEU5zKpoLyMHpDbcStSl6AXMepOHvKG5-ZbSgjK7Qb5Q. DOI: 10.5281/zenodo.20492145. [Access temporary limited only to Reviewers with the link shown above]. GPCRVS – a decision support system for GPCR virtual screening, based on machine learning and molecular docking, openly available on: https://gpcrvs.chem.uw.edu.pl.

Acknowledgments

We acknowledge Przemyslaw Osial and Faculty of Chemistry, University of Warsaw for providing the computational resources and web service maintaining.

Conflicts of Interest

The authors declare no known conflict of interest.

References

  1. Zundler, S.; Günther, C.; Kremer, A.E.; Zaiss, M.M.; Rothhammer, V.; Neurath, M.F. Gut Immune Cell Trafficking: Inter-Organ Communication and Immune-Mediated Inflammation. Nat. Rev. Gastroenterol. Hepatol. 2023, 20, 50–64. [Google Scholar] [CrossRef] [PubMed]
  2. Chimen, M.; Apta, B.H.R.; Mcgettrick, H.M. Introduction: T Cell Trafficking in Inflammation and Immunity. Methods Mol. Biol. 2017, 1591, 73–84. [Google Scholar] [CrossRef] [PubMed]
  3. Giovenzana, A.; Codazzi, V.; Pandolfo, M.; Petrelli, A. T Cell Trafficking in Human Chronic Inflammatory Diseases. iScience 2024, 27. [Google Scholar] [CrossRef] [PubMed]
  4. Song, B.; Guo, W.; He, Y.; Yao, X.; Sun, J.; Wang, S. Targeting Immune Cell Migration as Therapy for Inflammatory Disease: A Review. Front. Immunol. 2025, 16–2025. [Google Scholar] [CrossRef] [PubMed]
  5. Takeda, A.; Sasaki, N.; Miyasaka, M. The Molecular Cues Regulating Immune Cell Trafficking. Proc. Jpn. Acad. Ser. B Phys. Biol. Sci. 2017, 93, 183–195. [Google Scholar] [CrossRef] [PubMed]
  6. Latek, D.; Prajapati, K.; Dragan, P.; Merski, M.; Osial, P. GPCRVS - AI-Driven Decision Support System for GPCR Virtual Screening. Int. J. Mol. Sci. 2025, 26. [Google Scholar] [CrossRef] [PubMed]
  7. Bachelerie, F.; Ben-Baruch, A.; Burkhardt, A.M.; Combadiere, C.; Farber, J.M.; Graham, G.J.; Horuk, R.; Sparre-Ulrich, A.H.; Locati, M.; Luster, A.D.; et al. International Union of Basic and Clinical Pharmacology. LXXXIX. Update on the Extended Family of Chemokine Receptors and Introducing a New Nomenclature for Atypical Chemokine Receptors. Pharmacol. Rev. 2014, 66, 1–79. [Google Scholar] [CrossRef] [PubMed]
  8. Hughes, C.E.; Nibbs, R.J.B. A Guide to Chemokines and Their Receptors. FEBS J. 2018, 285, 2944–2971. [Google Scholar] [CrossRef] [PubMed]
  9. Proudfoot, A.E.I. Chemokine Receptors: Multifaceted Therapeutic Targets. Nat. Rev. Immunol. 2002, 2, 106–115. [Google Scholar] [CrossRef] [PubMed]
  10. Jung, H.; Bischof, A.; Ebsworth, K.; Ertl, L.; Schall, T.; Charo, I. Combination Therapy of Chemokine Receptor Inhibition plus PDL-1 Blockade Potentiates Anti-Tumor Effects in a Murine Model of Breast Cancer. j. Immunother. Cancer 2015, 3(P227), 2051-1426-3-S2-P227. [Google Scholar] [CrossRef]
  11. Sanford, D.E.; Belt, B.A.; Panni, R.Z.; Mayer, A.; Deshpande, A.D.; Carpenter, D.; Mitchem, J.B.; Plambeck-Suess, S.M.; Worley, L.A.; Goetz, B.D.; et al. Inflammatory Monocyte Mobilization Decreases Patient Survival in Pancreatic Cancer: A Role for Targeting the CCL2/CCR2 Axis. Clin. Cancer Res. 2013, 19, 3404–3415. [Google Scholar] [CrossRef] [PubMed]
  12. Figueras, A.; Alsina-Sanchís, E.; Lahiguera, Á.; Abreu, M.; Muinelo-Romay, L.; Moreno-Bueno, G.; Casanovas, O.; Graupera, M.; Matias-Guiu, X.; Vidal, A.; et al. A Role for CXCR4 in Peritoneal and Hematogenous Ovarian Cancer Dissemination. Mol. Cancer Ther. 2018, 17, 532–543. [Google Scholar] [CrossRef] [PubMed]
  13. Yoshimura, T.; Oppenheim, J.J. Chemokine-like Receptor 1 (CMKLR1) and Chemokine (C–C Motif) Receptor-like 2 (CCRL2); Two Multifunctional Receptors with Unusual Properties. Exp. Cell Res. 2011, 317, 674–684. [Google Scholar] [CrossRef] [PubMed]
  14. Graham, K.L.; Zhang, J.V.; Lewén, S.; Burke, T.M.; Dang, T.; Zoudilova, M.; Sobel, R.A.; Butcher, E.C.; Zabel, B.A. A Novel CMKLR1 Small Molecule Antagonist Suppresses CNS Autoimmune Inflammatory Disease. PLoS ONE 2014, 9, e112925. [Google Scholar] [CrossRef] [PubMed]
  15. Wang, D.; Mahmud, I.; Thakur, V.S.; Tan, S.K.; Isom, D.G.; Lombard, D.B.; Gonzalgo, M.L.; Kryvenko, O.N.; Lorenzi, P.L.; Tcheuyap, V.T.; et al. GPR1 and CMKLR1 Control Lipid Metabolism to Support the Development of Clear Cell Renal Cell Carcinoma. Cancer Res. 2024, 84, 2141–2154. [Google Scholar] [CrossRef] [PubMed]
  16. Rourke, J.L.; Dranse, H.J.; Sinal, C.J. CMKLR1 and GPR1 Mediate Chemerin Signaling through the RhoA/ROCK Pathway. Mol. Cell. Endocrinol. 2015, 417, 36–51. [Google Scholar] [CrossRef] [PubMed]
  17. Schioppa, T.; Sozio, F.; Barbazza, I.; Scutera, S.; Bosisio, D.; Sozzani, S.; Del Prete, A. Molecular Basis for CCRL2 Regulation of Leukocyte Migration. Front. Cell Dev. Biol. 2020, 8. [Google Scholar] [CrossRef] [PubMed]
  18. Ricklin, D.; Hajishengallis, G.; Yang, K.; Lambris, J.D. Complement: A Key System for Immune Surveillance and Homeostasis. Nat. Immunol. 2010. [Google Scholar] [CrossRef] [PubMed]
  19. Li, X.X.; Clark, R.J.; Woodruff, T.M. C5aR2 Activation Broadly Modulates the Signaling and Function of Primary Human Macrophages. J. Immunol. 2020, 205, 1102–1112. [Google Scholar] [CrossRef] [PubMed]
  20. Kildsgaard, J.; Hollmann, T.J.; Matthews, K.W.; Bian, K.; Murad, F.; Wetsel, R.A. Cutting Edge: Targeted Disruption of the C3a Receptor Gene Demonstrates a Novel Protective Anti-Inflammatory Role for C3a in Endotoxin-Shock. J. Immunol. 2000, 165, 5406–5409. [Google Scholar] [CrossRef] [PubMed]
  21. Ruocco, A.; Sirico, A.; Novelli, R.; Iannelli, S.; Van Breda, S.V.; Kyburz, D.; Hasler, P.; Aramini, A.; Amendola, P.G. The Role of C5a-C5aR1 Axis in Bone Pathophysiology: A Mini-Review. Front. Cell Dev. Biol. 2022, 10. [Google Scholar] [CrossRef] [PubMed]
  22. Wang, R.; Lu, B.; Gerard, C.; Gerard, N.P. Disruption of the Complement Anaphylatoxin Receptor C5L2 Exacerbates Inflammation in Allergic Contact Dermatitis. J. Immunol. 2013, 191, 4001–4009. [Google Scholar] [CrossRef]
  23. Zhang, T.; Wu, K.; Ma, N.; Wei, L.; Garstka, M.; Zhou, W.; Li, K. The C5a/C5aR2 Axis Promotes Renal Inflammation and Tissue Damage. JCI Insight 5 e134081. [CrossRef] [PubMed]
  24. Shadid, A.; Hok, K.D.; Domozhirov, A.Y.; Weng-Mills, T.; Doursout, M.-F.; Banda, N.K.; Restrepo, M.I.; Shivshankar, P. Enigmatic Roles of Complement Anaphylatoxin Signaling in Health and Disease. Immune Netw. 2025, 25, e32. [Google Scholar] [CrossRef] [PubMed]
  25. Boike, L.; Henning, N.J.; Nomura, D.K. Advances in Covalent Drug Discovery. Nat. Rev. Drug Discov. 2022, 21, 881–898. [Google Scholar] [CrossRef] [PubMed]
  26. Dragan, P.; Latek, D. The Two-Sided Impact of Beta-Adrenergic Receptor Ligands on Inflammation. Curr. Opin. Physiol. 2024, 41, 100779. [Google Scholar] [CrossRef]
  27. Eberhardt, J.; Santos-Martins, D.; Tillack, A.F.; Forli, S. AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings. J. Chem. Inf. Model. 2021, 61, 3891–3898. [Google Scholar] [CrossRef] [PubMed]
  28. Yang, Y.; Yao, K.; Repasky, M.P.; Leswing, K.; Abel, R.; Shoichet, B.K.; Jerome, S.V. Efficient Exploration of Chemical Space with Docking and Deep Learning. J. Chem. Theory Comput. 2021, 17, 7106–7119. [Google Scholar] [CrossRef] [PubMed]
  29. Daina, A.; Michielin, O.; Zoete, V. SwissTargetPrediction: Updated Data and New Features for Efficient Prediction of Protein Targets of Small Molecules. Nucleic Acids Res. 2019, 47, W357–W364. [Google Scholar] [CrossRef] [PubMed]
  30. Velloso, J.P.L.; Ascher, D.B.; Pires, D.E.V. pdCSM-GPCR: Predicting Potent GPCR Ligands with Graph-Based Signatures. Bioinform. Adv. 2021, 1, vbab031. [Google Scholar] [CrossRef] [PubMed]
  31. Liu, W.; Li, X.; Hang, B.; Wang, P. EnGCI: Enhancing GPCR-Compound Interaction Prediction via Large Molecular Models and KAN Network. BMC Biol. 2025, 23, 136. [Google Scholar] [CrossRef] [PubMed]
  32. Zhang, H.; Fan, H.; Wang, J.; Hou, T.; Saravanan, K.M.; Xia, W.; Kan, H.W.; Li, J.; Zhang, J.Z.H.; Liang, X.; et al. Revolutionizing GPCR–Ligand Predictions: DeepGPCR with Experimental Validation for High-Precision Drug Discovery. Brief. Bioinform. 2024, 25, bbae281. [Google Scholar] [CrossRef] [PubMed]
  33. Wang, Y.; Boadu, F.; Cheng, J. MPBind: A Multitask Protein Binding Site Predictor Using Protein Language Models and Equivariant GNNs. Bioinformatics 2025, 41, btaf589. [Google Scholar] [CrossRef] [PubMed]
  34. Garcia-Seyda, N.; Song, S.; Seveau de Noray, V.; David-Broglio, L.; Matti, C.; Artinger, M.; Dupuy, F.; Biarnes-Pelicot, M.; Valignat, M.-P.; Legler, D.F.; et al. Naive T Lymphocytes Chemotax Long Distance to CCL21 but Not to a Source of Bioactive S1P. iScience 2023, 26, 107695. [Google Scholar] [CrossRef] [PubMed]
  35. Wu, W.; Bao, Z.; Zhu, K.; Song, D.; Yang, W.; Luo, J.; Zheng, J.; Shao, G.; Huang, J. XCR1: A Promising Prognostic Marker That Pinpoints Targeted and Immune-Based Therapy in Hepatocellular Carcinoma. Heliyon 2024, 10, e31968. [Google Scholar] [CrossRef] [PubMed]
  36. Tondello, C.; Bender, C.; Golden, G.J.; Puppe, D.; Blickberndt, E.; Bayer, M.; Buchmann, G.K.; Pfeilschifter, J.; Bachmann, M.; Hintermann, E.; et al. The XCL1/XCR1 Axis Is Upregulated in Type 1 Diabetes and Aggravates Its Pathogenesis. JCI Insight 2025, 10. [Google Scholar] [CrossRef] [PubMed]
  37. Zanoni, M.; Aventurato, Í.K.; Hunter, J.; Sucupira, M.C.A.; Diaz, R.S. Uniquely Altered Transcripts Are Associated with Immune Preservation in HIV Infection. PLoS ONE 2017, 12, e0169868. [Google Scholar] [CrossRef] [PubMed]
  38. Lei, Y.; Takahama, Y. XCL1 and XCR1 in the Immune System. Microbes Infect. 2012, 14, 262–267. [Google Scholar] [CrossRef] [PubMed]
  39. Tian, Q.; Yan, Z.; Guo, Y.; Chen, Z.; Li, M. Inflammatory Role of CCR1 in the Central Nervous System. Neuroimmunomodulation 2024, 31, 173–182. [Google Scholar] [CrossRef] [PubMed]
  40. Zhang, P.; Dairaghi, D.J.; Jaen, J.C.; Powers, J.P. Chapter Ten - Recent Advances in the Discovery and Development of CCR1 Antagonists. In Annual Reports in Medicinal Chemistry; Desai, M.C., Ed.; Academic Press, 2013; Vol. 48, pp. 133–147. [Google Scholar]
  41. Metzemaekers, M.; Gouwy, M.; Proost, P. Neutrophil Chemoattractant Receptors in Health and Disease: Double-Edged Swords. Cell Mol. Immunol. 2020, 17, 433–450. [Google Scholar] [CrossRef] [PubMed]
  42. Gao, W.; Topham, P.S.; King, J.A.; Smiley, S.T.; Csizmadia, V.; Lu, B.; Gerard, C.J.; Hancock, W.W. Targeting of the Chemokine Receptor CCR1 Suppresses Development of Acute and Chronic Cardiac Allograft Rejection. J. Clin. Invest 2000, 105, 35–44. [Google Scholar] [CrossRef] [PubMed]
  43. Guo, S.; Zhang, Q.; Guo, Y.; Yin, X.; Zhang, P.; Mao, T.; Tian, Z.; Li, X. The Role and Therapeutic Targeting of the CCL2/CCR2 Signaling Axis in Inflammatory and Fibrotic Diseases. Front. Immunol. 2025, 15. [Google Scholar] [CrossRef] [PubMed]
  44. Yuan, J. CCR2: A Characteristic Chemokine Receptor in Normal and Pathological Intestine. Cytokine 2023, 169, 156292. [Google Scholar] [CrossRef] [PubMed]
  45. Zheng, Y.; Qin, L.; Zacarías, N.V.O.; de Vries, H.; Han, G.W.; Gustavsson, M.; Dabros, M.; Zhao, C.; Cherney, R.J.; Carter, P.; et al. Structure of CC Chemokine Receptor 2 with Orthosteric and Allosteric Antagonists. Nature 2016, 540, 458–461. [Google Scholar] [CrossRef] [PubMed]
  46. Miura, K.; Yang, L.; van Rooijen, N.; Ohnishi, H.; Seki, E. Hepatic Recruitment of Macrophages Promotes Nonalcoholic Steatohepatitis through CCR2. Am. J. Physiol.-Gastrointest. Liver Physiol. 2012, 302, G1310–G1321. [Google Scholar] [CrossRef] [PubMed]
  47. De Lema, G.P.; Maier, H.; Franz, T.J.; Escribese, M.; Chilla, S.; Segerer, S.; Camarasa, N.; Schmid, H.; Banas, B.; Kalaydjiev, S.; et al. Chemokine Receptor Ccr2 Deficiency Reduces Renal Disease and Prolongs Survival in MRL/Lpr Lupus-Prone Mice. J. Am. Soc. Nephrol. 2005, 16, 3592–3601. [Google Scholar] [CrossRef] [PubMed]
  48. Elsner, J.; Escher, S.E.; Forssmann, U. Chemokine Receptor Antagonists: A Novel Therapeutic Approach in Allergic Diseases. Allergy 2004, 59, 1243–1258. [Google Scholar] [CrossRef] [PubMed]
  49. Filippone, R.T.; Dargahi, N.; Eri, R.; Uranga, J.A.; Bornstein, J.C.; Apostolopoulos, V.; Nurgali, K. Potent CCR3 Receptor Antagonist, SB328437, Suppresses Colonic Eosinophil Chemotaxis and Inflammation in the Winnie Murine Model of Spontaneous Chronic Colitis. Int. J. Mol. Sci. 2022, 23, 7780. [Google Scholar] [CrossRef] [PubMed]
  50. Falcone, F.H.; Zillikens, D.; Gibbs, B.F. The 21st Century Renaissance of the Basophil? Current Insights into Its Role in Allergic Responses and Innate Immunity. Exp. Dermatol. 2006, 15, 855–864. [Google Scholar] [CrossRef] [PubMed]
  51. Dixon, H.; Blanchard, C.; deSchoolmeester, M.L.; Yuill, N.C.; Christie, J.W.; Rothenberg, M.E.; Else, K.J. The Role of Th2 Cytokines, Chemokines and Parasite Products in Eosinophil Recruitment to the Gastrointestinal Mucosa during Helminth Infection. Eur. J. Immunol. 2006, 36, 1753–1763. [Google Scholar] [CrossRef] [PubMed]
  52. Romagnani, P.; De Paulis, A.; Beltrame, C.; Annunziato, F.; Dente, V.; Maggi, E.; Romagnani, S.; Marone, G. Tryptase-Chymase Double-Positive Human Mast Cells Express the Eotaxin Receptor CCR3 and Are Attracted by CCR3-Binding Chemokines. Am. J. Pathol. 1999, 155, 1195–1204. [Google Scholar] [CrossRef] [PubMed]
  53. Bonecchi, R.; Bianchi, G.; Bordignon, P.P.; D’Ambrosio, D.; Lang, R.; Borsatti, A.; Sozzani, S.; Allavena, P.; Gray, P.A.; Mantovani, A.; et al. Differential Expression of Chemokine Receptors and Chemotactic Responsiveness of Type 1 T Helper Cells (Th1s) and Th2s. J. Exp. Med. 1998, 187, 129–134. [Google Scholar] [CrossRef] [PubMed]
  54. Blanchard, C.; Wang, N.; Stringer, K.F.; Mishra, A.; Fulkerson, P.C.; Abonia, J.P.; Jameson, S.C.; Kirby, C.; Konikoff, M.R.; Collins, M.H.; et al. Eotaxin-3 and a Uniquely Conserved Gene-Expression Profile in Eosinophilic Esophagitis. J. Clin. Invest 2006, 116, 536–547. [Google Scholar] [CrossRef] [PubMed]
  55. Viney, J.M.; Andrew, D.P.; Phillips, R.M.; Meiser, A.; Patel, P.; Lennartz-Walker, M.; Cousins, D.J.; Barton, N.P.; Hall, D.A.; Pease, J.E. Distinct Conformations of the Chemokine Receptor CCR4 with Implications for Its Targeting in Allergy. J. Immunol. 2014, 192, 3419–3427. [Google Scholar] [CrossRef] [PubMed]
  56. Yoshie, O. CCR4 as a Therapeutic Target for Cancer Immunotherapy. Cancers 2021, 13, 5542. [Google Scholar] [CrossRef] [PubMed]
  57. Bogacka, J.; Pawlik, K.; Ciapała, K.; Ciechanowska, A.; Mika, J. CC Chemokine Receptor 4 (CCR4) as a Possible New Target for Therapy. Int. J. Mol. Sci. 2022, 23, 15638. [Google Scholar] [CrossRef] [PubMed]
  58. Balistreri, C.R.; Caruso, C.; Grimaldi, M.P.; Listì, F.; Vasto, S.; Orlando, V.; Campagna, A.M.; Lio, D.; Candore, G. CCR5 Receptor. Ann. N. Y. Acad. Sci. 2007, 1100, 162–172. [Google Scholar] [CrossRef] [PubMed]
  59. Bambal, V.; Kodihalli Basavaraj, P.; Vanaja, K. Chemokine Receptor Type-5: A Key Regulator of Immunity, Disease Pathogenesis, and Emerging Therapeutic Target. Inflammopharmacol 2025, 33, 4477–4498. [Google Scholar] [CrossRef] [PubMed]
  60. Ajuebor, M.N.; Carey, J.A.; Swain, M.G. CCR5 in T Cell-Mediated Liver Diseases: What’s Going On? J. Immunol. 2006, 177, 2039–2045. [Google Scholar] [CrossRef] [PubMed]
  61. Sullivan, J.S. CC Chemokine Receptor 5 and Renal-Transplant Survival. The Lancet 2001, 358, 1269–1270. [Google Scholar] [CrossRef] [PubMed]
  62. Meitei, H.T.; Jadhav, N.; Lal, G. CCR6-CCL20 Axis as a Therapeutic Target for Autoimmune Diseases. Autoimmun. Rev. 2021, 20, 102846. [Google Scholar] [CrossRef] [PubMed]
  63. Ito, T.; Carson, W.F.; Cavassani, K.A.; Connett, J.M.; Kunkel, S.L. CCR6 as a Mediator of Immunity in the Lung and Gut. Exp. Cell Res. 2011, 317, 613–619. [Google Scholar] [CrossRef] [PubMed]
  64. Gómez-Melero, S.; Caballero-Villarraso, J. CCR6 as a Potential Target for Therapeutic Antibodies for the Treatment of Inflammatory Diseases. Antibodies 2023, 12, 30. [Google Scholar] [CrossRef] [PubMed]
  65. Krzysiek, R.; Lefevre, E.A.; Bernard, J.; Foussat, A.; Galanaud, P.; Louache, F.; Richard, Y. Regulation of CCR6 Chemokine Receptor Expression and Responsiveness to Macrophage Inflammatory Protein-3α/CCL20 in Human B Cells. Blood 2000, 96, 2338–2345. [Google Scholar] [CrossRef]
  66. Bruchard, M.; Geindreau, M.; Perrichet, A.; Truntzer, C.; Ballot, E.; Boidot, R.; Racoeur, C.; Barsac, E.; Chalmin, F.; Hibos, C.; et al. Recruitment and Activation of Type 3 Innate Lymphoid Cells Promote Antitumor Immune Responses. Nat. Immunol. 2022, 23, 262–274. [Google Scholar] [CrossRef] [PubMed]
  67. Comerford, I.; Harata-Lee, Y.; Bunting, M.D.; Gregor, C.; Kara, E.E.; McColl, S.R. A Myriad of Functions and Complex Regulation of the CCR7/CCL19/CCL21 Chemokine Axis in the Adaptive Immune System. Cytokine Growth Factor Rev. 2013, 24, 269–283. [Google Scholar] [CrossRef] [PubMed]
  68. Förster, R.; Davalos-Misslitz, A.C.; Rot, A. CCR7 and Its Ligands: Balancing Immunity and Tolerance. Nat. Rev. Immunol. 2008, 8, 362–371. [Google Scholar] [CrossRef] [PubMed]
  69. Legler, D.F.; Uetz-von Allmen, E.; Hauser, M.A. CCR7: Roles in Cancer Cell Dissemination, Migration and Metastasis Formation. Int. J. Biochem. Cell Biol. 2014, 54, 78–82. [Google Scholar] [CrossRef] [PubMed]
  70. Han, L.; Zhang, L. CCL21/CCR7 Axis as a Therapeutic Target for Autoimmune Diseases. Int. Immunopharmacol. 2023, 121, 110431. [Google Scholar] [CrossRef] [PubMed]
  71. Yabe, R.; Shimizu, K.; Shimizu, S.; Azechi, S.; Choi, B.-I.; Sudo, K.; Kubo, S.; Nakae, S.; Ishigame, H.; Kakuta, S.; et al. CCR8 Regulates Contact Hypersensitivity by Restricting Cutaneous Dendritic Cell Migration to the Draining Lymph Nodes. Int. Immunol. 2015, 27, 169–181. [Google Scholar] [CrossRef] [PubMed]
  72. Kvist Reimer, M.; Brange, C.; Rosendahl, A. CCR8 Signaling Influences Toll-Like Receptor 4 Responses in Human Macrophages in Inflammatory Diseases. Clin. Vaccine Immunol. 2011, 18, 2050–2059. [Google Scholar] [CrossRef] [PubMed]
  73. Mitsuyama, E.; Kunori, Y.; Kamimura, T.; Kaminuma, O. Functional Chemokine Receptors in Allergic Diseases: Is CCR8 a Novel Therapeutic Target? Mini-Rev. Med. Chem. 2006. [Google Scholar] [CrossRef] [PubMed]
  74. Ghosh, S.; Elder, A.; Guo, J.; Mani, U.; Patane, M.; Carson, K.; Ye, Q.; Bennett, R.; Chi, S.; Jenkins, T.; et al. Design, Synthesis, and Progress toward Optimization of Potent Small Molecule Antagonists of CC Chemokine Receptor 8 (CCR8). J. Med. Chem. 2006, 49, 2669–2672. [Google Scholar] [CrossRef] [PubMed]
  75. Diethelm, S.; Remeň, L.; Aissaoui, H.; Leroy, C.; Chavanton-Arpel, A.; Docci, L.; Seeland, S.; Lehembre, F.; Corminboeuf, O. Discovery of CCR8 Antagonist IDOR-1136-5177 for the Treatment of Cancer. ChemMedChem 2025, 20, e202400968. [Google Scholar] [CrossRef] [PubMed]
  76. Villarreal, D.O.; L’Huillier, A.; Armington, S.; Mottershead, C.; Filippova, E.V.; Coder, B.D.; Petit, R.G.; Princiotta, M.F. Targeting CCR8 Induces Protective Antitumor Immunity and Enhances Vaccine-Induced Responses in Colon Cancer. Cancer Res. 2018, 78, 5340–5348. [Google Scholar] [CrossRef] [PubMed]
  77. Saito, M.; Tanaka, T.; Asano, T.; Nakamura, T.; Yanaka, M.; Handa, S.; Komatsu, Y.; Harigae, Y.; Tateyama, N.; Nanamiya, R.; et al. C8Mab-2: An Anti-Mouse C–C Motif Chemokine Receptor 8 Monoclonal Antibody for Immunocytochemistry. Monoclon. Antib. Immunodiagn. Immunother. 2022, 41, 115–119. [Google Scholar] [CrossRef] [PubMed]
  78. Tiffany, H.L.; Lautens, L.L.; Gao, J.-L.; Pease, J.; Locati, M.; Combadiere, C.; Modi, W.; Bonner, T.I.; Murphy, P.M. Identification of CCR8: A Human Monocyte and Thymus Receptor for the CC Chemokine I-309. J. Exp. Med. 1997, 186, 165–170. [Google Scholar] [CrossRef] [PubMed]
  79. Huang, Y.; Wang, D.; Wang, X.; Zhang, Y.; Liu, T.; Chen, Y.; Tang, Y.; Wang, T.; Hu, D.; Huang, C. Abrogation of CC Chemokine Receptor 9 Ameliorates Ventricular Remodeling in Mice after Myocardial Infarction. Sci. Rep. 2016, 6, 32660. [Google Scholar] [CrossRef] [PubMed]
  80. Wu, X.; Sun, M.; Yang, Z.; Lu, C.; Wang, Q.; Wang, H.; Deng, C.; Liu, Y.; Yang, Y. The Roles of CCR9/CCL25 in Inflammation and Inflammation-Associated Diseases. Front. Cell Dev. Biol. 2021, 9. [Google Scholar] [CrossRef] [PubMed]
  81. Wurbel, M.-A.; Malissen, M.; Guy-Grand, D.; Meffre, E.; Nussenzweig, M.C.; Richelme, M.; Carrier, A.; Malissen, B. Mice Lacking the CCR9 CC-Chemokine Receptor Show a Mild Impairment of Early T- and B-Cell Development and a Reduction in T-Cell Receptor Γδ+ Gut Intraepithelial Lymphocytes. Blood 2001, 98, 2626–2632. [Google Scholar] [CrossRef] [PubMed]
  82. Martínez-Ríos, J.; López-Pacheco, C.P.; García-Zepeda, E.A.; Soldevila, G. CCR9 Shapes the Immune Microenvironment of Colorectal Cancer Modulating the Balance between Intratumoral CD8+ T Cell and FoxP3+ Helios+ Treg Subpopulations. PLoS ONE 2025, 20, e0321930. [Google Scholar] [CrossRef] [PubMed]
  83. Willuveit, A.L.; Stefanini, A.C.B.; Matozo, T.; Marti, L.C.C. CCR10: A Comprehensive Review of Its Function, Phylogeny, Role in Immune Cell Trafficking and Disease Pathogenesis. Front. Immunol. 2025, 16. [Google Scholar] [CrossRef] [PubMed]
  84. Liu, Q.; Li, A.; Tian, Y.; Wu, J.D.; Liu, Y.; Li, T.; Chen, Y.; Han, X.; Wu, K. The CXCL8-CXCR1/2 Pathways in Cancer. Cytokine Growth Factor Rev. 2016, 31, 61–71. [Google Scholar] [CrossRef] [PubMed]
  85. Sitaru, S.; Budke, A.; Bertini, R.; Sperandio, M. Therapeutic Inhibition of CXCR1/2: Where Do We Stand? Intern Emerg. Med. 2023, 18, 1647–1664. [Google Scholar] [CrossRef] [PubMed]
  86. Toya, S.; Struyf, S.; Huerta, L.; Morris, P.; Gavioli, E.; Minnella, E.M.; Cesta, M.C.; Allegretti, M.; Proost, P. A Narrative Review of Chemokine Receptors CXCR1 and CXCR2 and Their Role in Acute Respiratory Distress Syndrome. Eur. Respir. Rev. 2024, 33. [Google Scholar] [CrossRef] [PubMed]
  87. Ha, H.; Debnath, B.; Neamati, N. Role of the CXCL8-CXCR1/2 Axis in Cancer and Inflammatory Diseases. Theranostics 2017, 7, 1543–1588. [Google Scholar] [CrossRef] [PubMed]
  88. Wareing, M.D.; Lyon, A.B.; Lu, B.; Gerard, C.; Sarawar, S.R. Chemokine Expression during the Development and Resolution of a Pulmonary Leukocyte Response to Influenza A Virus Infection in Mice. J. Leukoc. Biol. 2004, 76, 886–895. [Google Scholar] [CrossRef] [PubMed]
  89. Liao, M.; Liu, Y.; Yuan, J.; Wen, Y.; Xu, G.; Zhao, J.; Cheng, L.; Li, J.; Wang, X.; Wang, F.; et al. Single-Cell Landscape of Bronchoalveolar Immune Cells in Patients with COVID-19. Nat. Med. 2020, 26, 842–844. [Google Scholar] [CrossRef] [PubMed]
  90. Lacotte, S.; Brun, S.; Muller, S.; Dumortier, H. CXCR3, Inflammation, and Autoimmune Diseases. Ann. N. Y. Acad. Sci. 2009, 1173, 310–317. [Google Scholar] [CrossRef] [PubMed]
  91. Wijtmans, M.; Scholten, D.; Mooij, W.; Smit, M.J.; de Esch, I.J.P.; de Graaf, C.; Leurs, R. Exploring the CXCR3 Chemokine Receptor with Small-Molecule Antagonists and Agonists. In Chemokines: Chemokines and Their Receptors in Drug Discovery; Tschammer, N., Ed.; Springer International Publishing: Cham, 2015; pp. 119–185. ISBN 978-3-319-14060-5. [Google Scholar]
  92. Groom, J.R.; Luster, A.D. CXCR3 Ligands: Redundant, Collaborative and Antagonistic Functions. Immunol. Cell Biol. 2011, 89. [Google Scholar] [CrossRef] [PubMed]
  93. Bao, S.; Darvishi, M.; H Amin, A.; Al-Haideri, M.T.; Patra, I.; Kashikova, K.; Ahmad, I.; Alsaikhan, F.; Al-qaim, Z.H.; Al-Gazally, M.E.; et al. CXC Chemokine Receptor 4 (CXCR4) Blockade in Cancer Treatment. J. Cancer Res. Clin. Oncol. 2023, 149, 7945–7968. [Google Scholar] [CrossRef] [PubMed]
  94. Döring, Y.; Pawig, L.; Weber, C.; Noels, H. The CXCL12/CXCR4 Chemokine Ligand/Receptor Axis in Cardiovascular Disease. Front Physiol. 2014, 5, 212. [Google Scholar] [CrossRef] [PubMed]
  95. García-Cuesta, E.M.; Santiago, C.A.; Vallejo-Díaz, J.; Juarranz, Y.; Rodríguez-Frade, J.M.; Mellado, M. The Role of the CXCL12/CXCR4/ACKR3 Axis in Autoimmune Diseases. Front Endocrinol. 2019, 10, 585. [Google Scholar] [CrossRef] [PubMed]
  96. Peled, A.; Wald, O.; Burger, J. Development of Novel CXCR4-Based Therapeutics. Expert Opin. Investig. Drugs 2012. [Google Scholar] [CrossRef] [PubMed]
  97. Stumm, R.; Höllt, V. CXC Chemokine Receptor 4 Regulates Neuronal Migration and Axonal Pathfinding in the Developing Nervous System: Implications for Neuronal Regeneration in the Adult Brain. J. Mol. Endocrinol. 2007, 38, 377–382. [Google Scholar] [CrossRef] [PubMed]
  98. Kazanietz, M.G.; Durando, M.; Cooke, M. CXCL13 and Its Receptor CXCR5 in Cancer: Inflammation, Immune Response, and Beyond. Front. Endocrinol. 2019, 10. [Google Scholar] [CrossRef] [PubMed]
  99. Pan, Z.; Zhu, T.; Liu, Y.; Zhang, N. Role of the CXCL13/CXCR5 Axis in Autoimmune Diseases. Front. Immunol. 2022, 13. [Google Scholar] [CrossRef] [PubMed]
  100. Harrer, C.; Otto, F.; Radlberger, R.F.; Moser, T.; Pilz, G.; Wipfler, P.; Harrer, A. The CXCL13/CXCR5 Immune Axis in Health and Disease—Implications for Intrathecal B Cell Activities in Neuroinflammation. Cells 2022, 11, 2649. [Google Scholar] [CrossRef] [PubMed]
  101. Deng, L.; Chen, N.; Li, Y.; Zheng, H.; Lei, Q. CXCR6/CXCL16 Functions as a Regulator in Metastasis and Progression of Cancer. Biochim. Et. Biophys. Acta (BBA) -Rev. Cancer 2010, 1806, 42–49. [Google Scholar] [CrossRef] [PubMed]
  102. Bao, N.; Fu, B.; Zhong, X.; Jia, S.; Ren, Z.; Wang, H.; Wang, W.; Shi, H.; Li, J.; Ge, F.; et al. Role of the CXCR6/CXCL16 Axis in Autoimmune Diseases. Int. Immunopharmacol. 2023, 121, 110530. [Google Scholar] [CrossRef] [PubMed]
  103. Dai, Y.; Wang, J.; Jeong, H.-H.; Chen, W.; Jia, P.; Zhao, Z. Association of CXCR6 with COVID-19 Severity: Delineating the Host Genetic Factors in Transcriptomic Regulation. bioRxiv 2021, 2021.02.17, 431554. [Google Scholar] [CrossRef] [PubMed]
  104. Lounsbury, N. Advances in CXCR7 Modulators. Pharmaceuticals 2020, 13, 33. [Google Scholar] [CrossRef] [PubMed]
  105. Duval, V.; Alayrac, P.; Silvestre, J.-S.; Levoye, A. Emerging Roles of the Atypical Chemokine Receptor 3 (ACKR3) in Cardiovascular Diseases. Front Endocrinol. 2022, 13, 906586. [Google Scholar] [CrossRef] [PubMed]
  106. Meyrath, M.; Szpakowska, M.; Zeiner, J.; Massotte, L.; Merz, M.P.; Benkel, T.; Simon, K.; Ohnmacht, J.; Turner, J.D.; Krüger, R.; et al. The Atypical Chemokine Receptor ACKR3/CXCR7 Is a Broad-Spectrum Scavenger for Opioid Peptides. Nat. Commun. 2020, 11, 3033. [Google Scholar] [CrossRef] [PubMed]
  107. Li, T.; Li, X.; Kang, P.; Zhao, J. Exploring CX3CR1 as a Prognostic Biomarker and Immunotherapeutic Target in Sarcoma. Transl. Oncol. 2025, 53, 102283. [Google Scholar] [CrossRef] [PubMed]
  108. Wei, L.-M.; Cao, S.; Yu, W.-D.; Liu, Y.-L.; Wang, J.-T. Overexpression of CX3CR1 Is Associated with Cellular Metastasis, Proliferation and Survival in Gastric Cancer. Oncol. Rep. 2015, 33, 615–624. [Google Scholar] [CrossRef] [PubMed]
  109. Liu, P.; Liang, Y.; Jiang, L.; Wang, H.; Wang, S.; Dong, J. CX3CL1/Fractalkine Enhances Prostate Cancer Spinal Metastasis by Activating the Src/FAK Pathway. Int. J. Oncol. 2018, 53, 1544–1556. [Google Scholar] [CrossRef] [PubMed]
  110. Wakita, H.; Yanagawa, T.; Kuboi, Y.; Imai, T. E6130, a Novel CX3C Chemokine Receptor 1 (CX3CR1) Modulator, Attenuates Mucosal Inflammation and Reduces CX3CR1+ Leukocyte Trafficking in Mice with Colitis. Mol. Pharmacol. 2017. [Google Scholar] [CrossRef] [PubMed]
  111. Zhuang, Q.; Cheng, K.; Ming, Y. CX3CL1/CX3CR1 Axis, as the Therapeutic Potential in Renal Diseases: Friend or Foe? Curr. Gene Ther. 2018. [Google Scholar] [CrossRef] [PubMed]
  112. Luo, P.; Chu, S.; Zhang, Z.; Xia, C.; Et., A. Fractalkine/CX3CR1 Is Involved in the Cross-Talk between Neuron and Glia in Neurological Diseases. Brain Res. Bull. 2019. [Google Scholar] [CrossRef] [PubMed]
  113. Bachelerie, F.; Murphy, P.M.; Nibbs, R.J.B.; Nomiyama, H.; Oppenheim, J.J.; Power, C.A.; Proudfoot, A.E.I.; Rosenkilde, M.M.; Rot, A.; Sozzani, S.; et al. Chemokine Receptors in GtoPdb v.2025.3. GtoPdb CITE 2025, 2025. [Google Scholar] [CrossRef]
  114. Schmitt, M. Deep Learning vs. Gradient Boosting: Benchmarking State-of-the-Art Machine Learning Algorithms for Credit Scoring. arXiv 2022, arXiv:2205.10535. [Google Scholar]
  115. McEwen, A.E.; Tejura, M.; Fayer, S.; Starita, L.M.; Fowler, D.M. Multiplexed Assays of Variant Effect for Clinical Variant Interpretation. Nat. Rev. Genet. 2026, 27, 137–154. [Google Scholar] [CrossRef] [PubMed]
  116. Wang, D.; Villenave, R.; Stokar-Regenscheit, N.; Clevers, H. Human Organoids as 3D in Vitro Platforms for Drug Discovery: Opportunities and Challenges. Nat. Rev. Drug Discov. 2026, 25, 204–226. [Google Scholar] [CrossRef] [PubMed]
  117. Gaulton, A.; Bellis, L.J.; Bento, A.P.; Chambers, J.; Davies, M.; Hersey, A.; Light, Y.; McGlinchey, S.; Michalovich, D.; Al-Lazikani, B.; et al. ChEMBL: A Large-Scale Bioactivity Database for Drug Discovery. Nucleic Acids Res. 2012, 40, D1100–7. [Google Scholar] [CrossRef] [PubMed]
  118. van Westen, G.J.P.; Gaulton, A.; Overington, J.P. Chemical, Target, and Bioactive Properties of Allosteric Modulation. PLoS Comput Biol. 2014, 10, e1003559. [Google Scholar] [CrossRef] [PubMed]
  119. Szpakowska, M.; D’Uonnolo, G.; Luís, R.; Alonso Bartolomé, A.; Thelen, M.; Legler, D.F.; Chevigné, A. New Pairings and Deorphanization among the Atypical Chemokine Receptor Family - Physiological and Clinical Relevance. Front Immunol. 2023, 14, 1133394. [Google Scholar] [CrossRef] [PubMed]
  120. Reyes, N.; Benedetti, I.; Rebollo, J.; Correa, O.; Geliebter, J. Atypical Chemokine Receptor CCRL2 Is Overexpressed in Prostate Cancer Cells. J. BioMed Res. 2017, 33, 17–23. [Google Scholar] [CrossRef] [PubMed]
  121. Terven, J.; Cordova-Esparza, D.-M.; Romero-González, J.-A.; Ramírez-Pedraza, A.; Chávez-Urbiola, E.A. A Comprehensive Survey of Loss Functions and Metrics in Deep Learning. Artif. Intell. Rev. 2025, 58, 195. [Google Scholar] [CrossRef]
  122. Dragan, P.; Merski, M.; Wiśniewski, S.; Sanmukh, S.G.; Latek, D. Chemokine Receptors—Structure-Based Virtual Screening Assisted by Machine Learning. Pharmaceutics 2023, 15, 516. [Google Scholar] [CrossRef] [PubMed]
  123. Fleming, J.; Magana, P.; Nair, S.; Tsenkov, M.; Bertoni, D.; Pidruchna, I.; Lima Afonso, M.Q.; Midlik, A.; Paramval, U.; Žídek, A.; et al. AlphaFold Protein Structure Database and 3D-Beacons: New Data and Capabilities. J. Mol. Biol. 2025, 437, 168967. [Google Scholar] [CrossRef] [PubMed]
  124. Yang, J.; Yan, R.; Roy, A.; Xu, D.; Poisson, J.; Zhang, Y. The I-TASSER Suite: Protein Structure and Function Prediction. Nat. Methods 2015, 12, 7–8. [Google Scholar] [CrossRef] [PubMed]
  125. Pándy-Szekeres, G.; Munk, C.; Tsonkov, T.M.; Mordalski, S.; Harpsøe, K.; Hauser, A.S.; Bojarski, A.J.; Gloriam, D.E. GPCRdb in 2018: Adding GPCR Structure Models and Ligands. Nucleic Acids Res. 2018, 46, D440–D446. [Google Scholar] [CrossRef] [PubMed]
  126. Kim, D.E.; Chivian, D.; Baker, D. Protein Structure Prediction and Analysis Using the Robetta Server. Nucleic Acids Res. 2004, 32, W526–W531. [Google Scholar] [CrossRef]
  127. Schrödinger, L.; DeLano, W. PyMOL 2020. [CrossRef]
  128. Robin, X.; Turck, N.; Hainard, A.; Tiberti, N.; Lisacek, F.; Sanchez, J.-C.; Müller, M. pROC: An Open-Source Package for R and S+ to Analyze and Compare ROC Curves. BMC Bioinform. 2011, 12, 77. [Google Scholar] [CrossRef] [PubMed]
  129. R Core Team R: A Language and Environment for Statistical Computing 2024.
  130. Apel, A.-K.; Cheng, R.K.Y.; Tautermann, C.S.; Brauchle, M.; Huang, C.-Y.; Pautsch, A.; Hennig, M.; Nar, H.; Schnapp, G. Crystal Structure of CC Chemokine Receptor 2A in Complex with an Orthosteric Antagonist Provides Insights for the Design of Selective Antagonists. Structure 2019, 27, 427–438.e5. [Google Scholar] [CrossRef] [PubMed]
  131. Isaikina, P.; Tsai, C.-J.; Dietz, N.; Pamula, F.; Grahl, A.; Goldie, K.N.; Guixà-González, R.; Branco, C.; Paolini-Bertrand, M.; Calo, N.; et al. Structural Basis of the Activation of the CC Chemokine Receptor 5 by a Chemokine Agonist. Sci. Adv. 2021, 7, eabg8685. [Google Scholar] [CrossRef] [PubMed]
  132. Tan, Q.; Zhu, Y.; Li, J.; Chen, Z.; Han, G.W.; Kufareva, I.; Li, T.; Ma, L.; Fenalti, G.; Li, J.; et al. Structure of the CCR5 Chemokine Receptor–HIV Entry Inhibitor Maraviroc Complex. Science 2013, 341, 1387–1390. [Google Scholar] [CrossRef] [PubMed]
  133. Wasilko, D.J.; Johnson, Z.L.; Ammirati, M.; Che, Y.; Griffor, M.C.; Han, S.; Wu, H. Structural Basis for Chemokine Receptor CCR6 Activation by the Endogenous Protein Ligand CCL20. Nat. Commun. 2020, 11, 3031. [Google Scholar] [CrossRef] [PubMed]
  134. Wasilko, D.J.; Gerstenberger, B.S.; Farley, K.A.; Li, W.; Alley, J.; Schnute, M.E.; Unwalla, R.J.; Victorino, J.; Crouse, K.K.; Ding, R.; et al. Structural Basis for CCR6 Modulation by Allosteric Antagonists. Nat. Commun. 2024, 15, 7574. [Google Scholar] [CrossRef] [PubMed]
  135. Jaeger, K.; Bruenle, S.; Weinert, T.; Guba, W.; Muehle, J.; Miyazaki, T.; Weber, M.; Furrer, A.; Haenggi, N.; Tetaz, T.; et al. Structural Basis for Allosteric Ligand Recognition in the Human CC Chemokine Receptor 7. Cell 2019, 178, 1222–1230.e10. [Google Scholar] [CrossRef] [PubMed]
  136. Jiang, S.; Lin, X.; Wu, L.; Wang, L.; Wu, Y.; Xu, Z.; Xu, F. Unveiling the Structural Mechanisms of Nonpeptide Ligand Recognition and Activation in Human Chemokine Receptor CCR8. Sci. Adv. 2024, 10, eadj7500. [Google Scholar] [CrossRef] [PubMed]
  137. Oswald, C.; Rappas, M.; Kean, J.; Doré, A.S.; Errey, J.C.; Bennett, K.; Deflorian, F.; Christopher, J.A.; Jazayeri, A.; Mason, J.S.; et al. Intracellular Allosteric Antagonism of the CCR9 Receptor. Nature 2016, 540, 462–465. [Google Scholar] [CrossRef] [PubMed]
  138. Liu, K.; Wu, L.; Yuan, S.; Wu, M.; Xu, Y.; Sun, Q.; Li, S.; Zhao, S.; Hua, T.; Liu, Z.-J. Structural Basis of CXC Chemokine Receptor 2 Activation and Signalling. Nature 2020, 585, 135–140. [Google Scholar] [CrossRef] [PubMed]
  139. Jiao, H.; Pang, B.; Liu, A.; Chen, Q.; Pan, Q.; Wang, X.; Xu, Y.; Chiang, Y.-C.; Ren, R.; Hu, H. Structural Insights into the Activation and Inhibition of CXC Chemokine Receptor 3. Nat. Struct. Mol. Biol. 2024, 31, 610–620. [Google Scholar] [CrossRef] [PubMed]
  140. Jiao, H.; Pang, B.; Chiang, Y.-C.; Chen, Q.; Pan, Q.; Ren, R.; Hu, H. Structure Basis for the Modulation of CXC Chemokine Receptor 3 by Antagonist AMG487. Cell Discov. 2023, 9, 1–4. [Google Scholar] [CrossRef] [PubMed]
  141. Saotome, K.; McGoldrick, L.L.; Ho, J.-H.; Ramlall, T.F.; Shah, S.; Moore, M.J.; Kim, J.H.; Leidich, R.; Olson, W.C.; Franklin, M.C. Structural Insights into CXCR4 Modulation and Oligomerization. Nat. Struct. Mol. Biol. 2025, 32, 315–325. [Google Scholar] [CrossRef] [PubMed]
  142. Liu, Y.; Liu, A.; Li, X.; Liao, Q.; Zhang, W.; Zhu, L.; Ye, R.D. Cryo-EM Structure of Monomeric CXCL12-Bound CXCR4 in the Active State. Cell Rep. 2024, 43, 114578. [Google Scholar] [CrossRef] [PubMed]
  143. Wu, B.; Chien, E.Y.T.; Mol, C.D.; Fenalti, G.; Liu, W.; Katritch, V.; Abagyan, R.; Brooun, A.; Wells, P.; Bi, F.C.; et al. Structures of the CXCR4 Chemokine GPCR with Small-Molecule and Cyclic Peptide Antagonists. Science 2010, 330, 1066–1071. [Google Scholar] [CrossRef] [PubMed]
  144. Yen, Y.-C.; Schafer, C.T.; Gustavsson, M.; Eberle, S.A.; Dominik, P.K.; Deneka, D.; Zhang, P.; Schall, T.J.; Kossiakoff, A.A.; Tesmer, J.J.G.; et al. Structures of Atypical Chemokine Receptor 3 Reveal the Basis for Its Promiscuity and Signaling Bias. Sci. Adv. 2022, 8, eabn8063. [Google Scholar] [CrossRef] [PubMed]
  145. Lu, M.; Zhao, W.; Han, S.; Lin, X.; Xu, T.; Tan, Q.; Wang, M.; Yi, C.; Chu, X.; Yang, W.; et al. Activation of the Human Chemokine Receptor CX3CR1 Regulated by Cholesterol. Sci. Adv. 2022, 8, eabn8048. [Google Scholar] [CrossRef] [PubMed]
  146. Wang, J.; Chen, G.; Liao, Q.; Lyu, W.; Liu, A.; Zhu, L.; Du, Y.; Ye, R.D. Cryo-EM Structure of the Human Chemerin Receptor 1–Gi Protein Complex Bound to the C-Terminal Nonapeptide of Chemerin. Proc. Natl. Acad. Sci. 2023, 120, e2214324120. [Google Scholar] [CrossRef] [PubMed]
  147. Liu, A.; Liu, Y.; Chen, G.; Lyu, W.; Ye, F.; Wang, J.; Liao, Q.; Zhu, L.; Du, Y.; Ye, R.D. Structure of G Protein-Coupled Receptor GPR1 Bound to Full-Length Chemerin Adipokine Reveals a Chemokine-like Reverse Binding Mode. PLoS Biol. 2024, 22, e3002838. [Google Scholar] [CrossRef] [PubMed]
  148. Lin, X.; Zhao, L.; Cai, H.; Chang, X.; Tang, Y.; Luo, T.; Wu, M.; Yi, C.; Ma, L.; Chu, X.; et al. Structural Insights into the Distinct Ligand Recognition and Signaling of the Chemerin Receptors CMKLR1 and GPR1. Protein Cell 2025, 16, 381–385. [Google Scholar] [CrossRef] [PubMed]
  149. Liao, Q.; Chen, G.; Du, Y.; Ye, R.D. Cryo-EM Structure of C3aR-Gi-scFv16 Bound with E7 Peptide: 8ia8 2023. [CrossRef]
  150. Feng, Y.; Zhao, C.; Deng, Y.; Wang, H.; Ma, L.; Liu, S.; Tian, X.; Wang, B.; Bin, Y.; Chen, P.; et al. Mechanism of Activation and Biased Signaling in Complement Receptor C5aR1. Cell Res. 2023, 33, 312–324. [Google Scholar] [CrossRef] [PubMed]
  151. Liu, H.; Kim, H.R.; Deepak, R.N.V.K.; Wang, L.; Chung, K.Y.; Fan, H.; Wei, Z.; Zhang, C. Orthosteric and Allosteric Action of the C5a Receptor Antagonists. Nat. Struct. Mol. Biol. 2018, 25, 472–481. [Google Scholar] [CrossRef] [PubMed]
  152. Eberhardt, J.; Santos-Martins, D.; Tillack, A.F.; Forli, S. AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings. J. Chem. Inf. Model. 2021, 61, 3891–3898. [Google Scholar] [CrossRef] [PubMed]
  153. Latek, D.; Rutkowska, E.; Niewieczerzal, S.; Cielecka-Piontek, J. Drug-Induced Diabetes Type 2: In Silico Study Involving Class B GPCRs. PLoS ONE 2019, 14, 1–27. [Google Scholar] [CrossRef] [PubMed]
Preprints 233855 g001
Figure 1. Performance of DNNs and GBMs in the ligand activity prediction for test datasets. The performance of DNNs (TensorFlow) and GBMs (LightGBM) was compared for test datasets (80/10/10 random split), not included in either training or validation datasets. DNN and GBM results mostly overlap for best-performing models but are complementary for weak-performing models trained using limited or fuzzy datasets. Here, I (e.g. CCR3-I) means inactive (datasets with inhibitors), A (e.g. CCR3-A) means active (datasets with agonists), and M (e.g. CCR8-M) means modulators (datasets with positive and negative modulators).
Figure 1. Performance of DNNs and GBMs in the ligand activity prediction for test datasets. The performance of DNNs (TensorFlow) and GBMs (LightGBM) was compared for test datasets (80/10/10 random split), not included in either training or validation datasets. DNN and GBM results mostly overlap for best-performing models but are complementary for weak-performing models trained using limited or fuzzy datasets. Here, I (e.g. CCR3-I) means inactive (datasets with inhibitors), A (e.g. CCR3-A) means active (datasets with agonists), and M (e.g. CCR8-M) means modulators (datasets with positive and negative modulators).
Preprints 233855 g001
Preprints 233855 g002
Figure 2. Differences between experimental and predicted pChEMBL values for control dataset compounds. Differences between experimental, ChEMBL-derived pChEMBL values and pChEMBL values predicted by GBMs (charts on the left), DNNs (charts in the middle), and AutoDock Vina (charts on the right) were calculated. Results for allosteric modes were shown in Supplementary Figure S2.
Figure 2. Differences between experimental and predicted pChEMBL values for control dataset compounds. Differences between experimental, ChEMBL-derived pChEMBL values and pChEMBL values predicted by GBMs (charts on the left), DNNs (charts in the middle), and AutoDock Vina (charts on the right) were calculated. Results for allosteric modes were shown in Supplementary Figure S2.
Preprints 233855 g002
Preprints 233855 g003
Figure 3. Structural models of GPCR receptors involved in the immune cell trafficking and inflammation implemented in GPCRVS. Regions refined with MODELLER/Rosetta were marked in red in the models.
Figure 3. Structural models of GPCR receptors involved in the immune cell trafficking and inflammation implemented in GPCRVS. Regions refined with MODELLER/Rosetta were marked in red in the models.
Preprints 233855 g003
Table 1. An overview of chemokine receptors included in GPCRVS and their major disease associations.
Table 1. An overview of chemokine receptors included in GPCRVS and their major disease associations.
receptor endogenous ligands key functions major disease links Ref.
action migrated cells
XC RECEPTORS
XCR1 XCL1, 2 cross-presentation, T cell activation, self-tolerance type 1 conventional dendritic cells (cDC1s), cytotoxic T cells AUTOIMMUNE type 1 diabetes, RA, pulmonary sarcoidosis, Crohn's disease, acute rejection of kidney transplantation
CANCER breast cancer, non-small cell lung cancer, renal cell carcinoma, hepatocellular carcinoma
INFECTIOUS HIV		 [35,36,37,38]
CC RECEPTORS
CCR1 CCL3–9, 13–16, 23 migration and activation macrophages, monocytes, neutrophils NEUROLOGICAL MS, Alzheimer's disease, stroke
CANCER multiple myeloma, osteolytic bone disease
AUTOIMMUNE RA, transplant rejection		 [39,40,41,42]
CCR2 CCL2, 7, 8, 11, 13, 16, 24, 26 migration immature dendritic cells, T cells, monocytes, macrophages, NK cells CARDIOVASCULAR atherosclerosis, atherosclerotic cardiovascular disease
AUTOIMMUNE & INFLAMMATORY MS, asthma, IBD, infective colitis, neuropathic pain, SLE
METABOLIC & HEPATIC diabetic nephropathy, metabolic dysfunction-associated steatohepatitis, fibrotic diseases
CANCER colon-related metastasis		 [43,44,45,46,47]
CCR3 CCL2, 5, 7, 8, 11, 13–15, 18, 24, 26, 28; CXCL9–11 migration eosinophils, basophils, mast cells, Th2 cells ALLERGIC & ATOPIC allergic asthma, allergic rhinitis, AD, eosinophilic esophagitis
INFLAMMATORY chronic colitis
INFECTIOUS parasitic/helminth infections		 [48,49,50,51,52,53,54]
CCR4 CCL17, 22; CKLF1 migration Th2 cells, Treg cells, mast cells ALLERGIC & ATOPIC allergic asthma, allergic rhinitis, AD
AUTOIMMUNE autoimmune encephalitis, MS
CANCER adult T cell leukemia/lymphoma, cutaneous T cell lymphomas, breast cancer, renal cancer
NEUROLOGICAL painful neuropathy		 [55,56,57]
CCR5 CCL2–5, 7, 8, 11, 13, 14, 16 migration and activation Th1 cells, macrophages, NK cells, immature dendritic cells CARDIOVASCULAR atherosclerosis
NEUROLOGICAL Alzheimer's disease
INFECTIOUS HIV infection
AUTOIMMUNE & INFLAMMATORY RA, IBD, T cell-mediated liver diseases, transplant rejection
CANCER basal and HER2+ breast cancer, liver cancer		 [58,59,60,61]
CCR6 CCL20 migration dendritic cells, professional APCs, Th17 cells, Treg cells, CD4 T cells, B cells, ILC3s AUTOIMMUNE & INFLAMMATORY IBD, psoriasis, RA, MS, AD, SLE, glomerulonephritis, vitiligo, endometriosis, dry eye disease
RESPIRATORY asthma, chronic obstructive pulmonary disease
METABOLIC & HEPATIC diabetes, chronic pancreatitis, cholestatic and chronic liver diseases
INFECTIOUS SARS-CoV-2
CANCER hepatocellular carcinoma; colorectal, breast, prostate, ovarian, lung, pancreatic, cervical, and renal cancer		 [62,63,64,65,66]
CCR7 CCL19, 21 lymphoid organogenesis, lymphocyte egress, memory and regulatory T cell function, migration T cells, B cells, dendritic cells, thymocytes AUTOIMMUNE RA, Sjögren's disease, SLE, polymyositis, ankylosing spondylitis
RESPIRATORY asthma
CANCER T cell leukemia, breast cancer, melanoma, non-small cell lung cancer, non-Hodgkin’s lymphoma, stomach cancer, CRC		 [67,68,69,70]
CCR8 CCL1, 18 migration, immune regulation, thymic cell line apoptosis Th2 cells, Treg cells, monocytes ATOPIC AD
CANCER breast cancer, CRC, melanoma, T cell leukemia and lymphoma
RESPIRATORY & ALLERGIC COPD, allergic asthma, allergic contact dermatitis
AUTOIMMUNE MS		 [71,72,73,74,75,76,77,78]
CCR9 CCL25 intestinal homing and migration T cells, B cells, dendritic cells AUTOIMMUNE & INFLAMMATORY Crohn's disease, IBD, celiac disease, rheumatoid arthritis, graft-versus-host disease, hepatitis, asthma
CANCER CRC, T cell acute lymphoblastic leukemia
CARDIOVASCULAR myocardial infarction		 [79,80,81,82]
CCR10 CCL27, 28 skin and mucosal homing T cells, plasma cells AUTOIMMUNE & INFLAMMATORY AD, psoriasis, allergic contact dermatitis
CANCER melanoma, cutaneous T cell lymphoma, adult T cell leukemia/lymphoma, Hodgkin’s lymphoma, non-small cell lung cancer, glioblastoma, breast cancer, ovarian cancer		 [83]
CXC RECEPTORS
CXCR1 CXCL6, 8 migration and activation neutrophils, macrophages, mast cells RESPIRATORY COPD, asthma, acute lung injury
INFLAMMATORY sepsis, RA, reperfusion injury
CANCER breast cancer, CRC, ovarian cancer, pancreatic cancer		 [84,85,86,87]
CXCR2 CXCL1–3, 5–8 migration and angiogenesis neutrophils, macrophages, mast cells RESPIRATORY COPD, asthma, bronchiectasis, cystic fibrosis
AUTOIMMUNE & INFLAMMATORY psoriasis, IBD, RA
CARDIOVASCULAR atherosclerosis
CANCER CRC, lung cancer, breast cancer, pancreatic cancer, melanoma, prostate cancer		 [84,85,86,87]
CXCR3 CCL5, 7, 11, 13, 19, 20; CXCL9–11, 12⍺ angiogenesis inhibition, migration Th1 cells, NK cells, CD8 T cells AUTOIMMUNE & INFLAMMATORY MS, RA, SLE, IBD
INFECTIOUS influenza, SARS-CoV-2,
CANCER colon cancer, melanoma, renal cancer, lymphoma, breast cancer		 [88,89,90,91,92]
CXCR4 CXCL12 hematopoiesis, organogenesis, tissue repair, embryonic development, neuronal migration, trafficking hematopoietic stem cells, lymphocytes, neurons INFECTIOUS HIV
AUTOIMMUNE & INFLAMMATORY RA, SLE, WHIM syndrome, asthma
CARDIOVASCULAR atherosclerosis, myocardial infarction
CANCER breast cancer, prostate cancer, lung cancer, CRC, leukemia, lymphoma, kidney cancer, pancreatic cancer, brain cancers, ovarian cancer, melanomas, 		[93,94,95,96,97]
CXCR5 CXCL13 lymphoid follicle homing, germinal center reaction B cells, follicular T helper cells AUTOIMMUNE SLE, RA, Sjögren's disease, MS
CANCER follicular lymphoma, B cell lymphoma, breast cancer, prostate cancer		 [98,99,100]
CXCR6 CXCL16 migration T cells, NK cells, NKT cells, plasma cells, monocytes AUTOIMMUNE & INFLAMMATORY IBD, RA, psoriasis, MS
HEPATIC autoimmune hepatitis, liver fibrosis, metabolic dysfunction-associated steatohepatitis
INFECTIOUS SARS-CoV-2
CANCER non-small cell lung cancer, prostate cancer, breast cancer		 [101,102,103]
CXCR7 (ACKR3) CXCL11, 12⍺ chemokine scavenging, cardiovascular and neural development, migration hematopoietic stem cells, CARDIOVASCULAR coronary artery disease, post-ischemic stroke
NEUROLOGICAL Alzheimer’s disease, epilepsy
AUTOIMMUNE & INFLAMMATORY MS, RA
CANCER breast cancer, lung cancer, prostate cancer, pancreatic cancer, ovarian cancer, CRC, kidney cancer, liver cancer, glioma		 [104,105,106]
CX3C RECEPTORS
CX3CR1 CX3CL1 adhesion, migration, and neuroprotection monocytes, NK cells, dendritic cells, microglia CARDIOVASCULAR atherosclerosis, renal ischemia
NEUROLOGICAL Alzheimer's disease, MS, neuropathic pain
AUTOIMMUNE & INFLAMMATORY IBD, RA, myositis
CANCER gastric cancer, prostate cancer, sarcoma		 [107,108,109,110,111,112]
Additional information, including endogenous ligands, was sourced from the IUPHAR/BPS Guide to Pharmacology (www.guidetopharmacology.org) [113]. Abbreviations: AD, atopic dermatitis; COPD, chronic obstructive pulmonary disease; CRC, colorectal cancer; IBD, inflammatory bowel disease; HIV, human immunodeficiency virus; MS, multiple sclerosis; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus.
Table 2. Performance comparison of one-receptor ML models implemented in GPCRVS. The best results were bolded if GBMs and DNNs differed.
Table 2. Performance comparison of one-receptor ML models implemented in GPCRVS. The best results were bolded if GBMs and DNNs differed.
Receptors Number of compounds in the training dataset GBMs - RMSE for test datasets GBMs – accuracy for test datasets [%] DNNs - accuracy for test datasets [%]
CCR1 inactive 1030 2.13 42.1 52.6
CCR1 active 58 2.77 28.6 28.6
CCR2 inactive 2718 4.73 52.3 55.3
CCR2 active 153 0.697 70.6 64.7
CCR3 inactive 1225 1.25 60.3 62.5
CCR3 active 191 0.410 57.1 52.4
CCR4 inactive 1357 4.89 65.6 68.2
CCR4 active 83 0.366 55.6 55.6
CCR5 inactive 3065 3.68 58.9 68.9
CCR5 active 118 0.537 92.3 92.3
CCR6 inactive 321 4.71 36.1 61.1
Mixed CCR6 active 618 0.173 82.6 81.2
Mixed CCR7 inactive 5350 4.19 57.0 60.0
Mixed CCR7 active 337 0.487 73.0 70.3
CCR8 inactive, orthosteric 132 0.317 53.3 60.0
Mixed CCR8 active, orthosteric 491 0.179 87.3 85.5
CCR8 allosteric modulators 19 0.704 50.0 50.0
CCR9 inactive 297 0.233 81.8 78.8
Mixed CCR9 active 245 0.999 66.7 66.7
CCR10 inactive 126 0.175 78.6 100
Mixed CCR10 active 719 0.261 78.8 80.0
CXCR1 inactive 1119 6.08 71.8 65.3
CXCR1 active 71 0.616 50.0 62.5
CXCR2 inactive 1944 4.56 56.5 57.4
CXCR2 active 151 0.195 64.7 70.6
CXCR3 inactive 1372 0.956 59.5 56.9
CXCR3 active 184 0.419 57.1 66.7
CXCR4 inactive 1215 0.459 76.3 69.6
CXCR4 active 252 0.221 75.0 71.4
CXCR5 inactive 145 0.489 75.0 81.3
Mixed CXCR5 active 618 0.191 81.2 82.6
CXCR7 (ACKR3) inactive 217 0.317 50.0 62.5
CXCR7 (ACKR3) active 472 0.442 58.5 52.8
CCRL2 inactive 139 0.432 56.2 62.5
Mixed CCRL2 active 289 1.14 75.0 81.3
CMKLR1 inactive 314 0.0818 88.6 88.6
CMKLR1 active 44 0.383 60.0 40.0
CX3CR1 inactive 418 0.0122 95.7 97.8
CX3CR1 active 355 0.00 100.0 100
C3aR inactive 174 0.615 57.9 68.4
C3aR active 132 0.749 46.7 46.7
C5aR inactive 947 0.160 85.7 86.7
C5aR active 438 0.0349 93.9 93.9
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.