Random forest approach to predict genes correlated with adaptive responses to KRASG12C inhibitor

KRAS mutations represent one of the most common and clinically challenging oncogenic drivers in non–small cell lung cancer (NSCLC), a leading cause of cancer mortality. The recent development of KRAS inhibitors, particularly targeting KRAS G12C has transformed the therapeutic landscape. However, despite these advances, treatment benefits remain limited as most patients experience disease progression within months due to adaptive or acquired resistance mechanisms. The incomplete understanding of the adaptive and acquired resistance mechanisms restricts our progression toward the development of effective therapeutic intervention. Particularly, we questioned whether common, targetable molecular vulnerabilities exist across distinct KRAS G12C tumor contexts during the early adaptive window under therapeutic pressure, preceding the establishment of overt drug resistance. To address this, we employed a Random Forest to single-cell RNA sequencing data from three KRAS G12C mutant NSCLC cell lines, enabling systematic ranking of genes associated with proliferative versus quiescent cell-state annotations. Our analysis recapitulated established proliferation-associated genes at baseline and revealed distinct gene signatures emerging at 72 hours of treatment, enriched for translational control and other biological pathways. Notably, PHF19 and TMEM106C were consistently prioritized across cell lines and showed higher direct expression in the proliferating populations, supporting their designation as candidate correlates of the residual proliferative state during early adaptation. These findings define proliferation and differentiation as a potential early vulnerability in KRAS G12C mutant NSCLC and provide a conceptual framework for intercepting resistance development, requiring subsequent external and functional validation.