PATH Integrates Multimodal Data to Identify Candidate Therapeutic Targets in Clear Cell Renal Cell Carcinoma

Background: Moving Beyond Single-Data-Type Target Discovery
The network medicine-based machine-learning framework PATH integrated distinct molecular datasets to propose candidate therapeutic targets for clear cell renal cell carcinoma (ccRCC). In follow-up cell-line experiments, an inhibitor targeting ENO2 showed the clearest antiproliferative effect. The validation was limited to cellular responses and did not establish efficacy in patients.
Traditional therapeutic target discovery has relied on single-gene hypotheses or analyses of mutations and gene expression. However, cancers are highly heterogeneous across cells and involve intertwined, multilayered molecular networks, making it difficult to identify disease-sustaining interactions from individual gene changes alone. This motivated the search for tumor-cell vulnerabilities in ccRCC, where resistance to existing treatments also occurs.
Key Findings: Integrating Four Data Types and Testing in Cell Lines
The researchers integrated single-cell transcriptomics, bulk multi-omics, genome-wide CRISPR knockout screening, and protein–protein interaction (PPI) networks. The CRISPR data came from existing DepMap datasets, not a new screen conducted by the research team.
PATH does not select candidates based only on gene expression levels. Using 35 gene signatures, the researchers calculated network distance, neighboring-gene composition, centrality, and the effects of gene knockout on cell viability, among other measures, to assemble 139 features for each of 17,880 genes. The trained model distinguishes known drug targets from other genes and ranks candidates. An ensemble model achieved a mean area under the receiver operating characteristic curve (AUROC) of 0.9254 on a balanced test set not used for training. This measures target-classification performance; it is not the clinical success rate of new candidates.
The researchers applied criteria including tumor-cell specificity and inhibitor availability to the predicted rankings, then tested six shortlisted candidates in A498, 769-P, and 786-O cell lines. Reduced cell viability was observed with inhibitors targeting ENO2, LRRK2, SCARB1, HMOX1, and TGM2, while a LOX inhibitor showed no effect. Responses to HMOX1 and TGM2 inhibition varied by cell line.
In proliferation experiments, each inhibitor was applied at 10 μM for 72 hours. POMHEX, an ENO2-targeting inhibitor, almost completely suppressed proliferation. The mean reduction in proliferation with LRRK2 and SCARB1 inhibitors was about 45% for each. Because these results came from different drugs, the figures alone do not establish the relative merits of the targets themselves.
Implications and Outlook: Repurposing Potential and the Limits of Validation
LRRK2 also draws interest from a drug-repurposing perspective. However, the drug used in these kidney cancer cell experiments was LRRK2-IN-1. It should be distinguished from BIIB122, which the paper describes as a drug in Phase 3 clinical development for Parkinson’s disease. This study did not demonstrate BIIB122’s efficacy against kidney cancer or establish reductions in drug-development time or cost.
The researchers also evaluated a PATH model built separately with lung adenocarcinoma-specific data and obtained an AUROC of 0.923. This supports the potential to distinguish known targets in another cancer type. Whether PATH can be applied immediately to diverse, hard-to-treat diseases requires separate validation.
The model does not directly explain the mechanism by which target inhibition leads to a therapeutic effect. Incomplete PPI data, research bias, and uncertainty in classifying genes as known targets or non-targets also remain limitations. Follow-up experiments must establish the candidates’ causal roles and therapeutic potential.
The methods and results are available in the full Communications Chemistry paper.
The high cost and attrition rate of drug development underscore the need for more effective strategies for therapeutic target discovery. Here, we present a network medicine-based machine learning framework that integrates single-cell transcriptomics, bulk multi-omic profiles, genome-wide CRISPR perturbation screens, and protein-protein interaction networks to systematically prioritise disease-specific targets. Applied to clear cell renal cell carcinoma, the framework successfully recovered established targets and predicted five therapeutic candidates, with subsequent in vitro validation demonstrating that among these, ENO2 inhibition had the strongest anti-tumour effect, followed by LRRK2, a repurposing candidate with phase III Parkinson's disease inhibitors. The proposed approach advances target discovery by moving beyond single-feature, single-modality heuristics to a scalable, machine learning-driven strategy that is generalisable across diseases.
PATH presents an approach that kidney cancer researchers could use to narrow down candidates for follow-up experiments. For example, applying criteria such as tumor-cell specificity, expression in normal tissues, and inhibitor availability alongside high computational rankings could help select candidates for testing. In this study, that process yielded six candidates for cell-line experiments. The work connects existing CRISPR data and multi-omics, converted into network features, with experimental prioritization. However, it has not been validated as universal guidance for weighting different data types or as a tool for selecting patient treatments. Advancing candidates into therapeutic targets will require further assessment of the drugs’ target specificity, mechanisms of action, safety, and in vivo effects.