Interpretable Recurrent Neural Networks for Modeling Temporal Dynamics of Acquired Drug Resistance in Targeted Therapy
Loading...
Files
Date
2026-05-26
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Acquired drug resistance remains a fundamental obstacle in targeted cancer therapy, often arising from complex
temporal evolutionary dynamics within tumor cell populations. Traditional pharmacokinetic-pharmacodynamic
(PKPD) models struggle to capture nonlinear, long-term sequential dependencies, while standard recurrent
neural networks (RNNs) suffer from black-box opacity. In this paper, we propose IRNN-DR (Interpretable RNN
for Drug Resistance), a novel architecture combining Long Short-Term Memory (LSTM) units with an
attention-based temporal relevance mask and a learnable dynamical system backbone. Using longitudinal single-
cell resistance data from EGFR-mutant non-small cell lung cancer (NSCLC) cell lines treated with osimertinib,
we demonstrate that IRNN-DR achieves high predictive accuracy (AUC-ROC = 0.94) while providing explicit
interpretations of resistance-driving time points and genetic pathways. Our contribution includes a temporal
Shapley value module and a resistance trajectory clustering method. Results indicate that IRNN-DR can identify
early "critical windows" of resistance emergence, with potential clinical utility for adaptive therapy scheduling