AI-powered multi-omic signature to predict treatment response of patients with colorectal cancer liver metastasis.

Abstract

252 Background: Colorectal cancer (CRC) is the third leading cause of cancer-related deaths in North America with over 50% of CRC patients developing liver metastases (LM) and 90% will die from metastatic disease. CRCLM are categorized into two main histological growth patterns (HGP) lesions: Desmoplastic (DHGP) and Replacement (RHGP). We have previously published that HGPs display distinct patterns of vascularization, local invasion, growth and response to treatment. Resected CRCLM patients with predominantly DHGP metastasis (~45%) receiving anti-VEGF therapy and chemotherapy have more than double the 5-year overall survival compared to patients with RHGP (~55%) who have received the same treatment. We are therefore treating patients that will not respond and should in fact be managed differently. Currently, there are no available biomarkers or stratification tools that predict colorectal cancer liver metastasis response to therapy. Methods: In collaboration with My Intelligent Machines (MIMs’), we used a proprietary GI2 (Genetic Interaction Graph Inference) algorithm inferring genetic interactions. We developed an AI based pipeline to develop a multi-modality signature, based on blood test analysis, to identify CRCLM patients who will 1) respond to Angiogenic Inhibitor-based therapies and 2) monitor the development of drug resistance (non-responders). Results: We obtained transcriptomic (tumor tissue) as well as proteomic data (liquid biopsy: Extracellular vesicle cargo), linked to clinical data from 40 chemonaïve patients, and fed the data into MIMs platform (BioMark). The software was able to identify differentially expressed genes common to the top differentially enriched proteins leading to a robust liquid biopsy biomarker signature distinguishing DHGP and RHGP. Furthermore, our data demonstrates a unique multi-modality signature providing biological insight for therapeutic failures and/or success. Conclusions: We will use these signatures to not only predict response to treatment but also decipher the molecular mechanisms driving these two phenotypes and identify unique targets. It is expected that this knowledge will lead to optimization of current treatment strategies, allowing for a precision therapy approach to the management of metastatic disease and cost saving for the health care system.