Identification of Predictive Signatures for Osimertinib‐Associated Interstitial Lung Disease: Integrating Ensemble Machine Learning and Temporal Dynamics

Vollständiger Abstract

Worum geht es in dieser Arbeit?

ABSTRACT The shift toward chronic management of EGFR‐mutant lung cancer underscores the urgent need to identify molecular features potentially relevant to osimertinib‐associated interstitial lung disease (ILD). This study aimed to investigate candidate molecular mechanisms of this toxicity by integrating putative osimertinib target profiles with ILD‐associated transcriptomic networks. To prioritize high‐confidence candidate features and improve cross‐algorithm robustness, we implemented an ensemble machine learning framework across 113 model configurations, using SHapley Additive exPlanations (SHAP) to quantify feature importance. Our analysis prioritized a six‐gene signature (ADRB2, DCLK1, PTGS2, CHRM3, MMP1, and CDK1) with consistent discriminative performance (AUC > 0.70), with lower ADRB2 and higher DCLK1 expression showing the strongest contributions to classification toward the ILD group. To assess the pharmacological relevance of these genes, we analyzed an osimertinib‐treated PC9 cell‐line time‐series dataset, in which exposure to 1 μM osimertinib for 24, 48, and 96 h was associated with distinct gene‐specific temporal expression patterns. Furthermore, molecular docking yielded favorable Vina docking scores for all six candidate proteins (belew −7.0 kcal/mol), suggesting potential structural compatibility with osimertinib. Collectively, these findings prioritize candidate molecular features and pathways potentially relevant to osimertinib‐associated ILD and provide a hypothesis‐generating framework for future prospective validation in osimertinib‐treated clinical cohorts.