Presentation: 2025 ND EPSCoR Annual conference 

October 21, 2025, NDSU Memorial Union, Fargo, North Dakota

QSAR Modeling to Predict Anticancer Activity of Drug-like Compounds in Hepatocellular Carcinoma

Hepatocellular carcinoma (HCC), the predominant form of primary liver cancer, remains a major contributor to cancer-related deaths worldwide. Limited treatment options and frequent recurrence highlight the need for better therapies. Computational modeling can support early drug discovery by estimating compound activity before committing to costly laboratory work. We tested whether a statistically validated machine learning (ML)-based quantitative structure–activity relationship (QSAR) model could predict anticancer activity of small molecules in the Huh-7 hepatocellular carcinoma cell line. A dataset of compounds with reported IC₅₀ values in Huh-7 cells was compiled from the literature. Structures were drawn in ChemDraw, optimized in HyperChem, and descriptors generated in alvaDesc. Multiple linear regression (MLR) was applied as a supervised ML method to develop models in QSARINS. Robustness was assessed with leave-one-out and leave-many-out cross-validation, an external test set, Y-scrambling, and William’s plot analysis. The final model showed strong predictive performance (R²train = 0.896, Q²LOO = 0.848, Q²LMO = 0.826, R²ext = 0.826), with low RMSE for both training (0.108) and external (0.128) sets. The concordance correlation coefficient (CCC) for the external set was 0.888. Y-scrambling indicated no chance correlation, and William’s plot confirmed all compounds were within the domain. Descriptor analysis identified H0v and CATS2D_04_LL as the most impactful features. These findings suggest the developed QSAR model can be a reliable tool for preliminary virtual screening of organic compounds with potential anticancer activity in HCC cells. Such screening may help prioritize candidates earlier in drug discovery and reduce experimental workload in oncology.