STDrug enables spatially informed personalized drug repurposing from spatial transcriptomics.

Area of Science:

• Computational biology
• Genomics
• Drug discovery

Background:

• Current drug repurposing methods often ignore spatial tissue context, limiting their effectiveness.
• Single-cell RNA sequencing (scRNA-seq) lacks the ability to capture microenvironment-dependent drug responses.

Purpose of the Study:

• To present STDrug, a novel computational framework for patient-specific therapeutic prioritization.
• To integrate spatial transcriptomics, graph-based modeling, and multimodal learning for enhanced drug repurposing.

Main Methods:

• STDrug utilizes graph convolutional networks and coherent point drift to identify and align spatial domains.
• It employs an integrative scoring scheme combining gene signatures, reversal scores, and machine learning for drug prioritization.
• The framework models spatial domain interactions, drug efficacy, and toxicity for patient-level drug scores.

Main Results:

• STDrug demonstrates superior performance compared to existing methods in hepatocellular carcinoma and prostate cancer datasets.
• Achieved significantly improved predictive accuracy (AUCs=0.81-0.82) across patients.
• Validation with electronic health records and in vitro assays supports the translational relevance of identified drug candidates.

Conclusions:

• STDrug provides a generalizable framework for integrating spatial omics into therapeutic discovery.
• It advances spatially informed and personalized drug repurposing strategies.
• The approach enhances the scalability and accuracy of identifying effective therapeutics.