Pan-Cancer Transcriptomic Clustering Reveals Conserved Molecular Programs and Cluster-Associated Compound Signatures

Abstract

Background: Pan-cancer transcriptomic analyses can identify molecular programs shared across tumor types. This study performed integrative unsupervised clustering of RNA-seq data to characterize expression-defined tumor groups and associated biological programs. Materials and Methods: RNA-seq data from 11,296 TCGA primary tumor samples across 33 cancer types were analyzed using preprocessing, principal component analysis, exploratory k-means clustering, differential-expression analysis, pathway enrichment, DSigDB compound-signature enrichment, independent CCLE analysis, and exploratory PRISM assessment. Results: Six exploratory transcriptomic clusters showed variable relationships with tissue oforigin and were associated with immune activation, cell-cycle regulation, metabolic processes,extracellular matrix remodeling, developmental signaling, and stress-response programs.DSigDB identified non-directional compound-signature overlaps involving retinoic acidderivatives, HDAC inhibitors, glucocorticoids, metabolic modulators, and toxicogenomicagents. CCLE analysis identified a stable six-group partition with partial concordance to TCGAclusters, and PRISM provided limited compound-specific pharmacogenomic support. Conclusion: This study characterizes transcriptomic heterogeneity across cancers and identifiesbiological and compound-signature associations that require further transcriptomic, clinical,pharmacogenomic, and experimental validation.