Explainable Machine Learning Model to Prediction EGFR Mutation in Lung Cancer.

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Lung adenocarcinoma (LA) diagnosis relies on gene sequencing.
• Predicting epidermal growth factor receptor (EGFR) mutations is crucial for targeted therapy.
• Explainable artificial intelligence models can potentially aid in clinical decision-making.

Purpose of the Study:

• To develop and validate an explainable machine learning model for predicting EGFR mutations in lung adenocarcinoma.
• To identify key clinical features and blood markers predictive of EGFR mutation status.
• To compare the performance of various machine learning algorithms in EGFR mutation prediction.

Main Methods:

• Retrospective analysis of 7,413 lung adenocarcinoma patients.
• Application of multiple machine learning algorithms including Random Forest (RF) and XGBoost.
• Utilizing demographic data, personal history, and blood markers as input features.
• Employing Area Under the Receiver Operating Characteristic Curve (AUC) and SHapley Additive exPlanation (SHAP) for model evaluation and interpretability.

Main Results:

• The Random Forest (RF) model achieved the highest AUC of 0.771 for EGFR mutation prediction.
• XGBoost also demonstrated strong performance with an AUC of 0.740.
• Key predictors identified include smoking consumption, sex, cholesterol, age, and albumin globulin ratio.
• SHAP analysis provided insights into feature influence on model predictions.

Conclusions:

• Machine learning algorithms, particularly RF, can effectively predict EGFR mutations in lung adenocarcinoma.
• The developed models offer a promising, explainable approach to complement traditional diagnostic methods.
• AI-based prediction models have the potential to guide clinical diagnosis and therapeutic strategies for LA patients.