Evolutionary Ensemble Learning Algorithm to Modeling of Warfarin Dose Prediction for Chinese.

Area of Science:

• Pharmacogenomics
• Computational Biology
• Machine Learning

Background:

• Warfarin dosing requires precise prediction due to its narrow therapeutic index.
• Existing models often lack sufficient accuracy for diverse patient populations.

Purpose of the Study:

• To develop and validate an improved warfarin dose prediction model using evolutionary ensemble modeling (EEM).
• To assess the performance of EEM against conventional and machine learning methods.

Main Methods:

• EEM combines genetic programming (GP) to evolve base models and genetic algorithms for parameter optimization.
• A bagging technique integrates diverse base models for the final ensemble.
• A dataset of 289 Chinese patients was used for training, validation, and testing.

Main Results:

• The EEM model, incorporating features like age, weight, gender, CYP2C9, VKORC1, and amiodarone (M2+G group), demonstrated superior predictive performance.
• EEM achieved the highest R², highest 20%-p, and lowest error metrics (MAE, MSE, RMSE) on the test set.
• The model exhibited minimal R² decrease from training to testing, indicating good generalization.

Conclusions:

• The EEM method, particularly with the M2+G feature group, provides a robust and accurate warfarin dose prediction.
• This EEM approach shows significant potential for clinical application in personalized warfarin therapy.