Integrative Transcriptomic and Machine Learning Analysis of ecDNA-Associated Features for Studying Chemotherapy Resistance in TNBC

remains incompletely understood. This study investigates the contribution of ecDNA-associated molecular features to predictive chemotherapy resistance in TNBC. We analyzed RNA-seq data from 4T1 TNBC cells and 4T1 bulk tumors at different growth stages (1-, 3-, and 6-week) to identify differentially expressed ecDNA alterations. We then utilized molecular docking tools to predict ecDNA protein-drug interactions and employed machine learning (ML) models to predict ecDNA-associated therapeutic resistance. Our results revealed changes in global gene expression, including expression of ecDNA-associated genes, that continued over time, with significant molecular remodeling observed at six weeks. Additionally, we found gradual accumulation of mutations in ecDNA genes, which may have contributed to reduced drug binding affinity, indicating potential resistance. ML models generated stable, high-confidence classifications of resistant phenotypes, consistently identifying ecDNA burden and prevalence as dominant predictive features of drug resistance. Drug specific predictions further highlighted elevated resistance probabilities for paclitaxel and doxorubicin, whereas hydroxyurea, which depletes ecDNA, showed reduced resistance probabilities, indicating potential roles of ecDNA in chemoresistance. This study provides new insights into temporal remodeling of ecDNA within TNBC tumors over time and their potential association with drug resistance.