Machine learning based personalized drug response prediction for lung cancer patients

Area of Science:

• Oncology
• Computational Biology
• Pharmacogenomics

Background:

• Epidermal growth factor receptor (EGFR) mutations drive significant lung cancer mortality.
• Targeted therapies like EGFR-tyrosine kinase inhibitors (TKIs) improve outcomes but face resistance.
• Predicting individual patient response to EGFR-TKIs is crucial for effective lung cancer treatment.

Purpose of the Study:

• To develop a personalized drug response prediction model (PDRP) for EGFR-mutated lung cancer.
• To predict patient response to first-generation EGFR-TKIs (Gefitinib/Erlotinib).
• To integrate molecular and clinical data for enhanced prediction accuracy.

Main Methods:

• Utilized molecular dynamics (MD) simulations to model patient-specific EGFR mutation status.
• Extracted molecular-level geometric features of the drug-target binding site.
• Integrated geometric features, binding free energy, and clinical data into an XGBoost machine learning classifier.

Main Results:

• The PDRP model achieved high performance: 97.5% accuracy, 93% recall, 96.5% precision, and 94% F1-score for a 4-class prediction task.
• Modeling binding pocket geometry and binding free energy proved effective predictors of drug response.
• Clinical information showed minimal impact on the model's predictive performance.

Conclusions:

• The PDRP model offers a promising tool for personalized treatment planning in EGFR-mutated lung cancer.
• Integrating molecular dynamics simulations with machine learning enhances drug response prediction.
• The model's framework has potential for application in other cancer types.