Leveraging Network-Based Transcriptome Analysis from Mouse Tumor Models and Explainable Artificial Intelligence to Advance the Understanding of the Antitumor Activity of Lenvatinib

Background/Objectives: Understanding the mechanisms of drug response plays an essential role in predicting effects prior to drug administration and advancing personalized medicine by optimizing treatment strategies. This study aimed to identify gene combinations that can predict the antitumor activity of lenvatinib, which is a multi-targeted tyrosine kinase inhibitor. Methods: Cancer- and drug-response-related gene sets were identified by mapping gene expression profiles of previously reported syngeneic mouse tumor models onto a protein–protein interaction network and extracting subnetworks comprising nodes where high expression levels were clustered. The scores for these network modules were calculated using the expression data of mouse tumor models prior to drug administration. These scores were used to train a machine learning (ML) model of drug response to lenvatinib by narrowing down the parameter space using hepatocellular carcinoma patient-derived xenograft (HCC PDX) models acquired in this study. Results: Using this integrative framework, we identified several network modules including those involved in the nerve growth factor signaling pathway, Wnt signaling pathway, and interleukin signaling pathways, that were consistently prioritized as informative features across PDX models and human patient data from The Cancer Genome Atlas. Conclusions: These network modules exhibit biological functions that are linked to the known targets of lenvatinib in the cancer cells or the tumor microenvironment, highlighting their potential relevance as determinants of drug response.