Integrated transcriptomic, machine-learning, and molecular simulation analysis prioritizes isovitexin-HSP90AB1 and cajanone-SERPINE1 interactions in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) remains a major cause of cancer mortality, highlighting the need for computational strategies that integrate disease transcriptomics with compound-target prediction and molecular simulation. This study investigated potential anti-HCC mechanisms of Cajanus cajan constituents through an integrated workflow. HCC-associated differentially expressed genes from TCGA-LIHC and three Gene Expression Omnibus datasets (GSE144269, GSE94660, and GSE14520) were intersected with predicted targets of selected C. cajan compounds. Protein-protein interaction analysis, functional enrichment, supervised machine-learning models, external validation, survival analysis, molecular docking, density functional theory calculations, molecular dynamics (MD) simulations, and MM/GBSA free-energy estimation were then applied. The integrative analyses prioritized a nine-gene network and identified HSP90AB1 and ESR1 as survival-associated markers, whereas SERPINE1 showed a nonsignificant risk trend. LASSO, random forest, and support vector machine classifiers displayed strong discrimination in the external dataset (area under the receiver operating characteristic curve, 0.981, 0.948, and 0.981, respectively). Docking ranked isovitexin-HSP90AB1 and cajanone-SERPINE1 among the favorable compound-target pairs. Primary 500-ns MD simulations, independent 100-ns replicates, and MM/GBSA analyses supported persistent interactions and stable conformational behavior for both complexes. These findings prioritize two testable C. cajan compound-protein hypotheses for HCC and provide a computational basis for subsequent biochemical and cellular validation.