Xprediction: Explainable EGFR-TKIs response prediction based on drug sensitivity specific gene networks

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Drug sensitivity prediction is crucial for precision medicine.
• Existing methods often overlook complex genetic interactions in diseases like cancer.
• Many machine learning models for prediction suffer from the 'black box' problem, hindering interpretability.

Purpose of the Study:

• To propose a novel strategy, Xprediction, for explainable drug sensitivity prediction.
• To integrate sample-specific gene regulatory networks into predictive models.
• To identify crucial molecular interactions underlying drug sensitivity and resistance.

Main Methods:

• Estimation of sample-specific gene regulatory networks.
• Application of machine learning models (random forest, kernel SVM, deep neural network) for prediction.
• Development of a method to quantify the importance of molecular interactions for prediction interpretability.

Main Results:

• Xprediction effectively predicts drug sensitivity using gene regulatory networks.
• The method identifies key gene-gene interactions relevant to drug sensitivity.
• Performance was superior to methods relying solely on gene expression levels.
• Identified markers for EGFR-TKIs prediction were validated through literature.

Conclusions:

• Xprediction offers an explainable approach to drug sensitivity prediction.
• The strategy enhances understanding of complex molecular mechanisms in drug response.
• It holds potential for predicting drug resistance and sensitivity in cancer cells.