A Genome-Informed Functional Modeling Approach to Evaluate the Responses of Breast Cancer Patients to CDK4/6 Inhibitors-Based Therapies and Simulate Real-World Clinical Trials

Abstract

AbstractPURPOSEVaried therapeutic responses were observed among cancer patients receiving the same treatment regimen, highlighting the challenge of identifying patients most likely to benefit from a given therapy. Here, we present an artificial intelligence-based approach, called CDK4/6 inhibitor Response Model (CRM), to address the complexity of predicting patient responses to treatment by a certain clinical scene on CDK4/6 inhibitors (CDK4/6i).PATIENTS AND METHODSTo train the CRM, we transformed the genomic data of 980 breast cancer patients from the TCGA database into activity profiles of signaling pathways (APSP) by utilizing the modified Damage Assessment of Genomic Mutations (DAGM) algorithm. A scoring model was then established by random forest algorithm to classify the HR+/HER2− and HR−/HER2− breast cancer molecular subtypes by the differential APSP features between the two, which reasonably reflected the potential role played by CDK4/6 molecules in HR+/HER2− breast cancer cells. The effectiveness of CRM was then tested in a separate local patient cohort (n = 343) in Guangdong, China. Twin in-silico clinical trials (ICT) of previously disclosed clinical trials (NCT02246621,NCT02079636,NCT03155997,NCT02513394,NCT02675231) were performed to demonstrate the potential of CRM in generating concerted results as the real-world clinical outcomes.RESULTSThe CRM displayed high precision in classifying HR+/HER2− and HR−/HER2− breast cancer patients in both TCGA (AUC=0.9956) and local patient cohorts (AUC=0.9795). Significantly, the scores were distinct (p = 0.025) between CDK4/6i-treated patients with different responses. Breast cancer patients from different subtypes were grouped into five distinct populations based on the scores assigned by the CRM. From twin ICT, the CRM scores reflected the differential responses of patient groups to CDK4/6i-based therapies.CONCLUSIONThe CRM score showed not only a robust association to clinically observed CDK4/6i responses but also heterogenetic responses across subtypes. More than half of HR+/HER2+ patients may be benefited from CDK4/6i-based treatment. The CRM empowered us to conduct ICT on different types of cancer patients responding to CDK4/6i-based therapies. These findings showed the potential of CRM as the companioned ICT to guide CDK4/6i application in the clinical end. CRM-guided ICT could be a universal method to demonstrate drug sensitivity to various patients.