EM-HIV

EM-HIV predicts human immunodeficiency virus 1 protease (HIV-1 PR) cleavage sites using an ensemble learning approach to inform protease inhibitor design and characterization of enzymatic specificity.

Key Features:

• Ensemble Learning: Integrates multiple weak learners via an asymmetric bagging technique to improve predictive performance.
• Biased Support Vector Machine Classifiers: Uses biased SVM classifiers as the weak learners within the ensemble.
• Feature Extraction from Substrate Sequences: Extracts features from octamer substrate sequences using amino acid identities, chemical properties of amino acids, and variable-length coevolutionary patterns.
• Octamer Feature Vectors: Constructs comprehensive feature vectors for octamers to represent cleavage-site candidates.
• Handling Data Uncertainty: Addresses data imbalance and noisy labels, including false positive and false negative cleavage sites, to reduce dataset separation uncertainty during training and testing.

Scientific Applications:

• Protease Inhibitor Design: Predicts cleavage sites to guide development of HIV-1 protease inhibitors.
• Enzymatic Activity Characterization: Maps HIV-1 PR cleavage specificity to improve understanding of protease substrate recognition.
• Comparative Evaluation and Benchmarking: Provides reliable predictions that support evaluation across independent benchmark datasets.

Methodology:

EM-HIV applies an ensemble learning framework that integrates biased support vector machine classifiers using asymmetric bagging; it extracts features from octamer substrate sequences via amino acid identity coding, chemical-property coding, and variable-length coevolutionary pattern coding, and constructs octamer feature vectors while addressing data imbalance and noisy labels such as false positives and false negatives to reduce dataset separation uncertainty during training and testing.