Multi-model inference using mixed effects from a linear regression based genetic algorithm.

Area of Science:

• Biostatistics
• Computational Biology
• Genetics

Background:

• High-dimensional regression requires effective variable selection for continuous variable prediction.
• Clustered observations in training data pose challenges for standard variable selection methods.
• Mixed-effects modeling (MM) offers a potential extension but can be complex to implement.

Purpose of the Study:

• To develop an automated mixed-effects modeling extension (GA-MM-MMI) for variable selection using a genetic algorithm (GA) and multi-model inference (MMI).
• To apply and evaluate this method for predicting Raltegravir (RAL) resistance from genotype-phenotype data with clustered observations.

Main Methods:

• Developed GA-MM-MMI, integrating a genetic algorithm (GA) with mixed-effects modeling (MM) and multi-model inference (MMI) for automated variable selection.
• Optimized GA parameters for computational efficiency using tournament selection and cross-validation.
• Applied the method to a genotype-phenotype dataset for predicting Raltegravir resistance, using integrase mutations as covariates.

Main Results:

• Achieved high model performance (R2MM > 0.95) with optimized GA parameters.
• The GA-MM-MMI approach identified a parsimonious model (GA-MM-MMI TOP18) with 18 key mutations.
• Demonstrated superior predictive accuracy (R2) compared to GA-ordinary least squares (GA-OLS) and LASSO on unseen data.

Conclusions:

• GA-MM-MMI significantly improves variable selection performance on genotype-phenotype datasets with clustered observations.
• The automated nature of parameter setting makes the method broadly applicable to similar datasets.
• The developed approach offers a more interpretable and accurate predictive model for complex biological data.