Constructing Individualized Cancer Networks To Predict Gene Targets
The expansive, unique genetic heterogeneity of individual tumors necessitates increasingly individualized approaches to therapeutics if patient outcomes are to improve. Omic data allows for a detailed perspective on individualized tumors, but its dimensionality makes mechanistic understanding and therapeutic targeting difficult. Network biology has proven to be a crucial tool for systems wide understanding, and machine learning enables the fine tuning of predictive pipelines. Here we combine these two tools in an individualized approach to understanding cancer biology and individualizing treatment. We adapt the PERsonalized Mutation evaluaTOR framework (Weiskittel et al. 2021) for generating individualized networks to create cell line specific modules. We then measure the topological features of nodes (genes) in these networks and evaluate their ability to accurately predict the success of therapeutically targeting specific genes in specific patients. We iteratively train machine learning models using network topology measures as features and gene dependency measurements (CRISPR KO from DepMap Data Portal) as labels. After this training process, we will have a generalizable model for predicting the optimal cancer gene targets in individualized tumors thus enabling high definition precision medicine. We expect that individualized network-derived features will be effective predictors of gene dependency, underscoring the usefulness of these networks to advance precision oncology.