An Integrated In Silico and In Vivo Genetic Screening Approach to Identify Subtype-specific Hepatocellular Carcinoma Genetic Dependencies

University Of Pittsburgh At Pittsburgh, Pittsburgh PA

Investigators

Abstract

PROJECT SUMMARY/ABSTRACT Hepatocellular carcinoma (HCC) is the 5th leading cause of cancer worldwide and the 4th leading cause of cancer- related mortality. Patients who present with advanced disease have poor prognosis, and currently approved systemic therapies confer only modest survival benefit. While HCC is a heterogeneous disease with subtypes marked by distinct clinical and pathomolecular features, precision therapies are currently lacking. In recent years, with the emergence of clinical trials for HCC targeted therapy, tissue biopsies are being increasingly utilized, presenting a timely opportunity to develop novel precision therapeutic approaches. However, most known molecular drivers of HCC remain pharmacologically difficult-to-target. Identifying targetable dependencies (genes essential for cancer cell proliferation) for HCC subclasses based on pathomolecular features thus represents a promising therapeutic strategy. We have previously developed a deep learning model to predict genetic dependencies based on multi-omic tumor profiles, and my preliminary work suggests that this model can predict dependencies specific to mutational subsets of HCC patients. Further, our group has developed murine models representing subtypes of HCC according to the Hoshida classification system (S1, S2, and S3). Given my preliminary observations, I hypothesize that genetic dependencies of HCC demonstrate subtype specificity based on pathomolecular features. I will address this hypothesis using a hybrid in vivo and in silico approach via the following specific aims. Specific Aim 1 (in vivo): I will determine whether dependencies predicted via deep learning demonstrate subtype specificity using in vivo loss-of-function screening and subsequently quantify the effect on tumor burden with single gene knockout of the most specific dependency for each subtype. This will enable efficient identification of potentially targetable, subtype-specific dependencies in robust preclinical models. Specific Aim 2 (in silico): I will expand the scope of our existing deep learning model to predict genetic dependencies from hematoxylin and eosin (H&E) images, after which I will compare the abundance of predictive H&E features across Hoshida subtypes to determine the subtype specificity of these features. I expect this to reveal novel, subtype-specific H&E features which predict HCC dependencies. These related but distinct and independent aims will contribute to our understanding of subtype-specific vulnerabilities in HCC, laying the groundwork for a theranostic pipeline to identify targetable dependencies based on patient-specific pathomolecular features. Through hands-on experience, mentorship, clinical exposure, conferences, coursework, and other activities described in my training plan, the education supported through this fellowship will prepare me for an impactful career as a physician-scientist at the forefront of digital pathology and precision medicine efforts.