GAXGB: A Two-Stage Ensemble Framework Integrating Genetic Algorithms and XGBoost for Anti-HIV Peptide Prediction

Abstract

This study applied a computational approach to derive amino acid sequence features relevant to AIDS treatment. A predictive model, GAXGB, was developed to classify anti-HIV peptides based on their amino acid sequence characteristics. The model was built using a 2-stage learning procedure. In the first stage, features were extracted from amino acid sequences using 12 descriptors, and 120 baseline models were constructed with 10 different classifiers. In the second stage, these baseline models generated 120 predictive probability scores, which were used as input features. Various feature selection methods, including chi-square, ANOVA, mutual information, and a genetic algorithm, were employed to identify the most significant features for the final model. Subsequently, 10 classifiers were trained and evaluated. Performance evaluation showed that the GAXGB model, which combines genetic algorithm-based feature selection with an XGBoost classifier, achieved superior predictive accuracy. The model reached an accuracy of 90%, significantly outperforming other models that achieved approximately 80% accuracy. This approach offers a promising tool to accelerate the design and discovery of novel anti-HIV peptides for AIDS treatment.