Predicting Emergent Tumor Drug Resistance in Triple Negative Breast Cancer through Bayesian Tumor Phylogenies and Deep Learning*
Session Number
3
Advisor(s)
Dr. John Huelsenbeck, University of California, Berkeley
Location
A117
Discipline
Biology
Start Date
15-4-2026 2:15 PM
End Date
15-4-2026 3:00 PM
Abstract
Triple-negative breast cancer (TNBC) frequently develops resistance to chemotherapy through subclonal evolution, but most existing computational oncology methods do not fully account for this. They reconstruct clonal structure or identify resistance-associated features after treatment rather than forecast resistance accumulation on evolving tumor lineages. This project develops a Bayesian deep phylogenetic pipeline that combines phyddle-based inference of tumor topology and lineage-specific evolutionary parameters with continuous-trait modeling in RevBayes to estimate resistance accumulation over time in TNBC. Using longitudinal sequencing data from a TNBC cohort collected across various stages of chemotherapy, we develop a framework that treats resistance not as a binary state but as a stochastic trait evolving along branches of a phylogeny. This project works to integrate posterior phylogenetic structure with a predictive model of resistance accumulation through evolutionary time. Specifically, this work lies in translating posterior tree structure and inferred branch-rate information in R into deep-learning-guided forecasts of subclonal resistance dynamics in C++ and Python, thereby unifying tumor phylogenetic inference and continuous resistance modeling in a single framework for early identification of high-risk drug-resistant subclones. We are currently leveraging this framework to identify emergent vulnerabilities in tumor populations, as well as find novel drug targets.
Predicting Emergent Tumor Drug Resistance in Triple Negative Breast Cancer through Bayesian Tumor Phylogenies and Deep Learning*
A117
Triple-negative breast cancer (TNBC) frequently develops resistance to chemotherapy through subclonal evolution, but most existing computational oncology methods do not fully account for this. They reconstruct clonal structure or identify resistance-associated features after treatment rather than forecast resistance accumulation on evolving tumor lineages. This project develops a Bayesian deep phylogenetic pipeline that combines phyddle-based inference of tumor topology and lineage-specific evolutionary parameters with continuous-trait modeling in RevBayes to estimate resistance accumulation over time in TNBC. Using longitudinal sequencing data from a TNBC cohort collected across various stages of chemotherapy, we develop a framework that treats resistance not as a binary state but as a stochastic trait evolving along branches of a phylogeny. This project works to integrate posterior phylogenetic structure with a predictive model of resistance accumulation through evolutionary time. Specifically, this work lies in translating posterior tree structure and inferred branch-rate information in R into deep-learning-guided forecasts of subclonal resistance dynamics in C++ and Python, thereby unifying tumor phylogenetic inference and continuous resistance modeling in a single framework for early identification of high-risk drug-resistant subclones. We are currently leveraging this framework to identify emergent vulnerabilities in tumor populations, as well as find novel drug targets.