A Multi-Layered Proteogenomic Framework for Prioritization of Cell Surface Therapeutic Targets in Solid Tumors

1. Introduction

Target selection is a critical and rate-limiting step in the development of precision oncology therapeutics, particularly for modalities that rely on target accessibility at the cell surface, such as radioligand therapy (RLT), radioimmunotherapy (RIT), antibody–drug conjugates, and other antibody-based platforms [1,2]. The clinical success of these approaches depends on the identification of tumor-associated antigens that are abundantly expressed on the plasma membrane of cancer cells, exhibit limited expression in normal tissues, and present extracellular epitopes amenable to therapeutic binding and internalization. Failure to adequately address these criteria during early target discovery frequently leads to suboptimal efficacy or unacceptable off-target toxicity in later stages of development.

Despite major advances in genomics, transcriptomics, and proteomics, the systematic identification of clinically actionable cell surface targets in solid tumors remains challenging. Many discovery efforts rely predominantly on differential gene expression analyses or literature-driven prioritization, which do not reliably capture protein-level abundance, subcellular localization, or therapeutic accessibility [3,4]. In addition, transcriptomic signals often correlate poorly with surface protein expression, particularly for membrane proteins subject to post-transcriptional regulation, variable trafficking, or proteolytic processing [3,5]. As a result, candidate lists derived from single-layer analyses are frequently enriched for false positives and lack translational robustness.

From a therapeutic standpoint, the “ideal” cell surface target is rarely defined by a single parameter [1,6]. Instead, clinically viable targets reflect a balance between multiple, sometimes competing, criteria, including tumor-selective overexpression, restricted normal tissue distribution, stable membrane localization without extensive shedding, evidence of functional relevance in cancer biology, and feasibility for drug development. Importantly, translational considerations such as prior clinical validation, availability of biomarker data, and compatibility with specific therapeutic modalities are often evaluated late in the discovery process, increasing the risk of downstream attrition.

To address these challenges, there is a growing need for integrative discovery frameworks that combine quantitative proteomics with curated biological and clinical metadata, while explicitly accounting for uncertainty and dataset-specific variability [7,8]. Rather than relying on a single optimal ranking scheme, parallel filtering strategies can provide complementary perspectives on the data, enabling identification of candidates that are robust to parameter selection and cohort effects. Recurrent detection across independent strategies may therefore serve as an implicit robustness metric for target prioritization.

Recent pan-cancer proteogenomic studies have demonstrated the power of integrating tumor-level protein expression with genomic data to expand the landscape of candidate therapeutic targets. In particular, Savage et al. reported a large-scale analysis across multiple cancer types, establishing proteomics as a critical layer for target discovery beyond genomics and transcriptomics alone [7]. However, global target catalogs are not designed to address key translational constraints relevant to therapeutic modalities that require extracellular target accessibility, nor do they provide indication-specific prioritization or robustness assessment under alternative analytical assumptions. These considerations are particularly important for antibody-based therapies and radioligand therapy, where plasma membrane localization and normal tissue expression critically influence clinical feasibility.

Here, we present a multi-layered proteogenomic filtering pipeline designed to systematically identify and prioritize cell surface proteins for therapeutic targeting in solid tumors. The pipeline integrates quantitative proteomics from large-scale cancer cohorts with curated metadata on protein localization, normal tissue expression, expression in cancer cell lines, and clinical drugability. Eleven complementary filtering strategies were applied, followed by manual curation to confirm extracellular accessibility and composite scoring based on expression, localization, and translational relevance. Importantly, the approach emphasizes overlap and consistency across strategies rather than reliance on any single filtering criterion.

We apply this framework to metastatic colorectal cancer (mCRC) as a case study, a disease with high unmet clinical need and increasing interest in targeted radiopharmaceutical and antibody-based therapies. Using data from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) [3,4], Human Protein Atlas [9], Genotype-Tissue Expression project, Cancer Cell Line Encyclopedia, and curated drug development databases, we demonstrate that this multi-layered approach identifies both established colorectal cancer biomarkers and previously underexplored candidate targets. The resulting prioritization highlights GPRC5A as a top candidate while maintaining sensitivity to diverse protein classes and biological mechanisms.

Together, this study provides a scalable and generalizable framework for cell surface target discovery that explicitly integrates proteomic evidence with translational constraints. Beyond colorectal cancer, the modular design of the pipeline enables adaptation to other solid tumor indications and therapeutic modalities, supporting early-stage decision-making in precision oncology drug development.

2. Materials and Methods

Data Sources

The pipeline integrates multiple publicly available and proprietary datasets to enable comprehensive target discovery. Quantitative protein expression data were obtained from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) COAD and COADREAD cohorts, comprising colorectal adenocarcinoma and rectal cancer samples with matched normal tissue controls [3,4]. Protein abundance in normal tissues was assessed using data from the Human Protein Atlas (HPA) and the Genotype-Tissue Expression (GTEx) project [9,10]. RNA sequencing data for colorectal cancer cell lines were retrieved from the Cancer Cell Line Encyclopedia (DepMap) [11]. Clinical metadata describing drugability and biomarker status were sourced from Clarivate Drug Discovery Intelligence (CDDI, Cortellis platform; Clarivate Analytics), while information on the number of unique drugs, associated indications, and highest clinical development stage was extracted from Swiss-Prot [12].

Pipeline Architecture

The target discovery pipeline was designed as a modular, multi-stage framework integrating quantitative proteomics with curated biological and clinical metadata. The architecture was optimized to enable systematic prioritization of cell surface proteins while accommodating dataset-specific variability and uncertainty inherent to large-scale proteomic analyses. The pipeline consists of four core stages: (i) construction of an annotated union dataset, (ii) parallel multi-parameter filtering, (iii) manual curation for extracellular accessibility, and (iv) composite scoring and prioritization (Figure 1).

Annotated Union Dataset Construction

All datasets were merged into a unified, protein-level table using standardized protein identifiers. For each protein, the annotated union table included quantitative protein expression ranks and differential expression metrics derived from CPTAC COAD and COADREAD cohorts, subcellular localization annotations obtained from UniProt, Human Protein Atlas (HPA), and the Cell Surface Protein Atlas (CSPA), normal tissue expression levels from HPA and GTEx, expression in colorectal cancer cell lines from the Cancer Cell Line Encyclopedia (DepMap), and clinical development metadata from Clarivate Drug Discovery Intelligence (CDDI) and SwissProt.

To systematically assess therapeutic accessibility, a custom localization score was generated for each protein based on curated annotations indicating plasma membrane localization, transmembrane topology, and presence of extracellular domains. Secreted proteins and proteins exclusively annotated to intracellular compartments were flagged for exclusion in later steps.

Parallel Filtering Strategies

Rather than applying a single optimal filtering scheme, the pipeline implements multiple complementary filtering strategies in parallel (Table 1, Table 2 and Table 3). Each strategy represents a distinct combination of selection criteria targeting overexpression in metastatic disease, plasma membrane localization, minimal normal tissue expression, and expression in cancer cell lines. This design reduces dependence on any single parameter threshold and allows identification of candidates that are robust to methodological variability.

Filtering strategies were grouped conceptually into three categories: (i) rank-based strategies combining protein abundance, localization scores, and normal tissue penalties; (ii) differential expression-based strategies focused on upregulated plasma membrane proteins; and (iii) cohort-driven strategies applying stricter differential expression cut-offs within individual CPTAC cohorts. All strategies shared core architectural principles but differed in stringency and parameter weighting, enabling broad exploration of the target space while preserving biological relevance.

Each filtering strategy generated an independent shortlist of candidate proteins. Proteins detected across multiple strategies were considered more robust, reflecting consistency across analytical perspectives.

Manual Curation for Extracellular Accessibility

Candidates emerging from the automated filtering steps were subjected to manual curation to confirm extracellular accessibility and therapeutic relevance. This step focused on reviewing protein topology, domain structure, and annotation consistency to exclude proteins lacking extracellular epitopes despite partial membrane association. Manual curation was essential to eliminate false positives arising from incomplete or ambiguous database annotations and to ensure compatibility with antibody-based and radioligand-based therapeutic modalities [1,6].

Composite Scoring and Prioritization

Final prioritization of candidate targets was performed using a composite scoring framework integrating multiple quantitative and qualitative parameters. The composite score incorporated stage IV protein expression rank, localization score, penalties for normal tissue expression (HPA and GTEx), expression in colorectal cancer cell lines, and clinical development metrics reflecting drugability and prior biomarker use. Candidates identified by multiple filtering strategies were weighted more strongly, reflecting increased robustness. Weighting of individual parameters was intentionally kept conservative and balanced to avoid overfitting to any single dataset or feature, prioritizing robustness over optimization.

The resulting ranked list represents a balanced prioritization of candidate targets based on expression, accessibility, safety, and translational maturity, rather than optimization of any single criterion.

Validation Layers

To contextualize and illustrate the performance of the multi-layered target discovery pipeline, selected high-ranking candidate proteins were evaluated using several independent validation layers. These analyses were not intended as definitive experimental validation, but rather as orthogonal assessments addressing key translational risks, including expression robustness, localization accuracy, relevance beyond the discovery dataset, and prior clinical or biomarker evidence.

Protein Expression Visualization

Quantitative protein expression patterns were visualized across tumor stages using swarm plots of normalized protein abundance from the CPTAC COAD and COADREAD cohorts. These plots enabled assessment of stage-associated expression trends, inter-patient variability, and consistency between cohorts, with particular emphasis on overexpression in advanced and metastatic disease. Normal tissue reference samples were included where available to contextualize tumor-specific expression.

Transcriptomic Analysis

To assess concordance between protein-level signals and transcriptomic data, mRNA expression patterns for candidate targets were examined in colorectal cancer cell lines from the Cancer Cell Line Encyclopedia (DepMap) and patient tumor samples from The Cancer Genome Atlas (TCGA). Heatmaps were generated to visualize relative expression across samples and to evaluate whether proteomic signals were supported by transcriptional data, recognizing that discordance between RNA and protein expression may occur for membrane-associated proteins.

Immunohistochemistry and Subcellular Localization

Protein localization and tissue-level expression patterns were examined using immunohistochemistry data from the Human Protein Atlas (HPA). Emphasis was placed on confirming plasma membrane staining and differential expression between colorectal cancer tissue and normal colon, where available. This layer served as an independent qualitative assessment of extracellular accessibility and supported exclusion of candidates with ambiguous or predominantly intracellular localization.

Literature Mining and Biomarker Evidence

To evaluate prior evidence linking candidate targets to cancer biology and clinical investigation, systematic literature mining was performed using PubMed. The number of publications reporting associations with cancer, metastatic disease, and metastatic colorectal cancer was recorded. This analysis provided contextual insight into existing biological and clinical knowledge, without presupposing target validity or drugability.

Protein Expression in Other Cancer Types

To assess whether candidate targets exhibited broader oncologic relevance or tumor-type specificity, protein expression data from additional CPTAC cohorts were examined, including breast invasive carcinoma (BRCA; PDC000120), lung squamous cell carcinoma (LSCC; PDC000234), lung adenocarcinoma (LUAD; PDC000153), ovarian serous cystadenocarcinoma (OV; PDC000110), and pancreatic ductal adenocarcinoma (PDAC; PDC000270). Comparative analysis across indications helped contextualize colorectal cancer findings and informed prioritization based on potential therapeutic breadth or selectivity.

Clinical and Biomarker Annotation

Finally, candidate targets were evaluated using curated clinical metadata from Clarivate Drug Discovery Intelligence (CDDI) to determine whether they had previously been reported as biomarkers or explored in therapeutic development. Metrics included number of reported indications, cancer-specific usage, and highest development phase where applicable. This layer provided translational context and assisted in distinguishing well-characterized targets from emerging or underexplored candidates.

Collectively, these validation layers are intended to provide orthogonal context for target prioritization rather than functional validation, which will require dedicated experimental studies. Localization and accessibility assessments were particularly emphasized due to their importance for radiopharmaceutical and antibody-based targeting.

3. Results

3.1. Pipeline Performance Across Filtering Strategies

The multi-layered filtering pipeline was applied to proteomic and metadata from the CPTAC COAD and COADREAD cohorts [3,4] to identify cell surface proteins associated with metastatic colorectal cancer (mCRC) according to the workflow described in Figure 1. Eleven distinct filtering strategies were implemented, each varying in stringency for protein expression, localization, normal tissue expression and expression on cancer cell lines (Table 1, Table 2 and Table 3).

The number of candidate proteins identified per strategy ranged from 5 to 58 and the number of unique proteins identified for each strategy ranged from 0 to 29, reflecting differences in parameter thresholds and dataset-specific variability (Figure 2). Strategies 1–4, which combined protein rank and localization scores with normal tissue penalties, yielded between 10 and 58 candidates per iteration. Strategies 5–7, incorporating differential expression and manual curation for plasma membrane localization, produced 5–7 candidates. Cohort-driven strategies (8–11), which applied stricter log fold-change cut-offs to either COAD or COADREAD independently, identified 6–23 candidates after manual review.

Candidate targets are identified through multiple complementary filtering strategies applied in parallel, each combining criteria related to metastatic overexpression, plasma membrane localization, extracellular accessibility, and limited normal tissue expression. Manual curation is subsequently performed to confirm extracellular exposure and to exclude proteins with ambiguous or non therapeutically accessible localization. Final prioritization is achieved using a composite scoring framework that integrates expression, localization, safety related penalties, and translational relevance. Candidates supported by multiple filtering strategies are considered more robust and prioritized for downstream validation.

CPTAC=Clinical Proteomic Tumor Analysis Consortium; COAD=Colon Adenocarcinoma; COADREAD=Colon Adenocarcinoma and Rectal Adenocarincoma; CRC=Colorectal Cancer; BRCA=Breast Invasive Carcinoma; LSCC=Lung squamous cell carcinoma; LUAD=Lung Adenocarcinoma; OV=Ovarian Serous Cystadenocarcinoma; PDAC=Pancreatic Ductal Adenocarcinoma; CDDI=Cortellis Drug Discovery Intelligence; HPA=Human Protein Atlas; GTEX=The Genotype-Tissue Expression project; CSPA=Cell Surface Protein Atlas; CCLE=Cancer Cell Line Encyclopedia; GO=Gene Ontology; TCGA=The Cancer Genome Atlas.

3.2. Overlap and Robustness of Candidate Selection

To assess robustness, candidate lists from all strategies were compared. A union analysis revealed that only a subset of 43 proteins was identified by more than one strategy. A weighted scoring was implemented where candidates identified by the most stringent filtering strategies (4-10) were assigned a weight of 2 and candidates identified by less stringent strategies (1-3 and 11) were assigned a weight of 1. Figure 3 shows the resulting 48 proteins with a score of more than 1. GPRC5A emerged as the most robust candidate, appearing in 8 different strategies and all major strategy groups (2–4, 5–7, and 8–9). Other recurrent hits included IFITM1, SLC5A6, DPEP1 and LY75, each detected in 5 or more independent strategies (Figure 3).

The average pairwise overlap between filtering strategies was 0.11, indicating limited redundancy between individual approaches. Despite this low overlap, candidates detected across multiple strategies were strongly enriched for known colorectal cancer biomarkers. Specifically, FAP, CEACAM5, ITGAV, and ITGB4 were exclusively identified among multi-strategy candidates (4/39, 10.3%) and were absent among single-strategy hits (0/95), representing a significant enrichment (Fisher’s exact test, p = 0.0064). These findings support cross-strategy recurrence as a marker of biological and translational relevance.

The goal for this analysis was to find unique targets for radiopharmaceuticals. Strategy 5-11 did not contain criteria for exclusion of targets that were already exploited in this field. Therefore, a new round of selection was performed on the data in Figure 3 to remove the candidates where we could find that it had been used as a target for a therapeutic radiopharmaceutical agent. This analysis resulted in exclusion of ITGB4 [13], CEACAM5 [14] Bioorg Med Chem Lett, 2024], FAP [15], CD46 [16] and SLC7A5 [17]. Supplementary Figure 1 shows the resulting bar chart. Importantly, this exclusion was performed as a secondary analysis step and was not incorporated into the core pipeline, ensuring that initial target identification remained unbiased.

3.3. Composite Scoring and Prioritization

Final candidate ranking was based on a composite score integrating mean stage IV protein rank, localization score, normal tissue expression penalty, and protein expression in cancer cell lines. Table 4 shows the resulting Top-10 candidates. GPRC5A achieved the highest composite score (4.5), followed by SLC2A1 (4.0), CD47 (2.5), DPEP1 (1.9) and IFITM1 (1.9). These candidates represent diverse protein classes, including a G-protein coupled receptor, a solute carriers, an immune checkpoint regulator, an enzyme and an antiviral effector, highlighting the pipeline’s capacity to capture biologically distinct targets.

3.4. Validation Layers

To illustrate the translational potential of the pipeline, selected candidates were evaluated using multiple complementary validation layers. PubMed citation analysis showed that all top-ranked candidates have previously been reported in cancer-related contexts, with most also described in metastatic disease, supporting their relevance as established or emerging biomarkers (Figure 4). Among these, GPRC5A has been associated with adverse clinical outcomes and tumor progression across multiple cancer types, supporting its biological relevance as a therapeutic candidate [18]. In colorectal cancer specifically, GPRC5A expression has been reported to promote tumor growth and invasion through non-coding RNA–mediated regulatory mechanisms [19].

Additional top-ranked candidates were likewise supported by independent literature evidence. SLC2A1 is frequently overexpressed in colorectal cancer and has been linked to disease progression and immune modulation, while its plasma membrane localization and central role in tumor metabolism make it an attractive therapeutic target [20,21]. CD47 represents a well-established innate immune checkpoint in colorectal cancer, and its recovery among the top-ranked candidates provides an internal validation of the pipeline’s ability to capture clinically actionable cell-surface targets [22,23]. DPEP1 is a GPI-anchored surface enzyme with high specificity for colorectal cancer, where it has been shown to promote invasion and metastasis and to associate with adverse clinical outcome [24,25]. IFITM1 is overexpressed in metastatic colorectal cancer and functionally promotes invasion, and recent work has demonstrated its suitability as a surface imaging and targeting antigen in vivo [26,27].

Immunohistochemistry data from the Human Protein Atlas further verified plasma membrane localization for GPRC5A, SLC2A, CD47, DPEP1, IFITM1, DSC2 and IFITM3 in colorectal cancer tissues (Figure 5). In addition, protein expression analysis across other tumor types demonstrated that most of the top-10 ranked candidates were differentially expressed in multiple cancer indications (Figure 6). Finally, clinical annotation revealed that all top-ranked candidates have previously been used as biomarkers, including across different cancer types, further supporting their translational relevance (Figure 7).

3.5. Summary of Pipeline Robustness

The integration of proteogenomic data with curated metadata and iterative filtering strategies enabled the identification of high-confidence cell surface targets for mCRC. The recovery of known biomarkers alongside novel candidates demonstrates the validity and generalizability of the approach. The pipeline’s modular design allows adaptation to other solid tumor indications and therapeutic modalities.

4. Discussion

The identification of tumor-specific cell surface targets remains a major challenge in oncology drug development, particularly for modalities such as radioligand therapy (RLT) and radioimmunotherapy (RIT), where off-target toxicity can severely limit clinical applicability. Our study introduces a multi-layered proteogenomic filtering pipeline that addresses key limitations of existing approaches by integrating quantitative proteomics with curated metadata on protein localization, normal tissue expression, and clinical drugability.

Recent pan-cancer proteogenomic atlases have demonstrated the power of large-scale integration of proteomic and genomic data for expanding the landscape of candidate cancer drug targets, most notably the comprehensive analysis reported by Savage et al. [7]. While such efforts provide invaluable resources for global target enumeration, their objectives differ from the translational challenges addressed here. Our study focuses on indication-specific prioritization by explicitly incorporating extracellular accessibility, normal tissue expression penalties, and robustness across multiple complementary filtering strategies. This distinction is particularly relevant for antibody-based therapies and radiopharmaceuticals, where plasma membrane localization and on-target toxicity are dominant determinants of clinical feasibility [1,2].

Advantages of the Pipeline

The pipeline’s strength lies in its modular design and multi-parameter filtering strategy. By combining eleven complementary filtering strategies with manual curation, we achieved robust candidate selection while minimizing reliance on single-layer analyses that are prone to false positives. This is supported by the consistent recovery of clinically validated targets such as FAP and CEACAM5 across independent filtering strategies. In addition, candidates identified by multiple strategies were enriched for known cancer-associated surface proteins, suggesting that overlap across filters represents a practical proxy for biological robustness. Importantly, the identification of less-characterized candidates such as GPRC5A and SLC5A6 highlights the ability of the pipeline to move beyond canonical targets while retaining translational relevance.

Comparison with Existing Approaches

Traditional target discovery pipelines often prioritize candidates based on differential expression alone, which can lead to high false-positive rates and poor translational performance. This may partly explain the high attrition rate observed when targets identified through transcriptomic profiling alone are advanced toward therapeutic development. Our method incorporates additional layers of evidence, including plasma membrane localization and normal tissue expression penalties, which are critical for reducing on-target toxicity. Furthermore, the integration of clinical metadata enables prioritization of targets with established drugability, facilitating downstream development.

Limitations

Despite its strengths, the pipeline has limitations. First, the approach relies on the quality and completeness of underlying datasets. Differences in normalization and technology platforms between CPTAC cohorts introduced variability in differential expression analysis, necessitating separate cohort-driven strategies. Second, manual curation remains essential for confirming extracellular accessibility, as database annotations do not always provide sufficient resolution. This step, while improving accuracy, limits scalability and introduces subjectivity. Third, the current study focuses on target identification and prioritization and does not include functional validation of candidate targets in vitro or in vivo. Such experiments will be required to confirm target dependency, internalization properties, and therapeutic relevance.

Future Directions

Future iterations of the pipeline should focus on automation of curation steps through advanced protein topology prediction tools and machine learning-based annotation. Integration of additional omics layers, such as phosphoproteomics and mutational data, may further enhance target prioritization by linking expression patterns to functional relevance. Finally, expanding the pipeline to other solid tumor indications will validate its generalizability and support its application in diverse therapeutic contexts, including antibody-drug conjugates and bispecific antibodies.

Clinical and Translational Implications

The pipeline provides a systematic framework for identifying high-confidence targets for precision oncology. By structuring target prioritization around clinically relevant parameters such as tissue specificity and accessibility, this framework may support more efficient transition from discovery to preclinical validation, thereby reducing the number of non-viable targets entering early development pipelines. Its adaptability to different cancer types and therapeutic modalities positions it as a valuable tool for next-generation drug development. In particular, the prioritization of cell surface proteins with low normal tissue expression aligns closely with the requirements for radioligand therapy, where target accessibility and tumor specificity are critical determinants of therapeutic index. Although radioligand development was not performed in this study, the identified candidates provide a rational starting point for future ligand design and evaluation.

Among the prioritized candidates, GPRC5A is of particular interest given its reported overexpression in multiple epithelial malignancies and potential role in tumor progression. Similarly, SLC5A6 represents a transporter protein with limited characterization in cancer but favorable surface accessibility, making it an intriguing candidate for targeted delivery approaches. Further investigation of these targets may provide insight into both tumor biology and therapeutic vulnerability.

5. Conclusions

We developed and validated a multi-layered proteogenomic filtering pipeline for the systematic identification of cell-surface therapeutic targets in solid tumors. Using metastatic colorectal cancer (mCRC) as a case example, the pipeline integrates quantitative proteomics with curated metadata on protein localization, normal tissue expression, and clinical drugability to address key limitations of traditional target discovery approaches. Eleven complementary filtering strategies, combined with manual curation and composite scoring, enabled robust and reproducible candidate selection while explicitly incorporating translational constraints early in the discovery process.

Application of this framework consistently identified high-confidence targets, including GPRC5A, SLC2A1, CD47, DPEP1 and IFITM1, several of which are established biomarkers or emerging therapeutic candidates. Overlap analysis demonstrated that candidates detected across multiple independent strategies were enriched for known cancer-relevant surface markers, supporting the validity and robustness of the approach. In addition, the use of composite scoring provided a transparent and reproducible mechanism for prioritization, facilitating comparison of candidates across heterogeneous analytical dimensions and reducing reliance on single-parameter ranking.

By integrating multiple data layers and emphasizing robustness and therapeutic accessibility, this pipeline reduces false-positive target nomination and helps bridge the gap between discovery-stage proteogenomics and translational decision-making. Its modular design allows straightforward adaptation to other cancer indications and therapeutic platforms, including radioligand therapy, antibody–drug conjugates, and bispecific antibodies. Future development will focus on automating currently manual curation steps, incorporating additional omics layers such as phosphoproteomics and mutational data, and validating performance across diverse tumor types. Together, these enhancements are expected to improve scalability and further accelerate the translation of novel cell-surface targets into clinical applications.