catkern

catkern predicts HIV drug resistance from viral protein sequence data using weighted categorical kernel functions to model allele mixtures and residue-specific contributions to resistance.

Key Features:

• Weighted categorical kernel functions: Implements weighted categorical kernels tailored for sequence data with allele mixtures to capture categorical site states.
• Residue-specific weighting: Assigns different weights to protein residues to reflect variable contributions to drug resistance.
• Kernel comparison: Compares categorical kernels (Overlap and Jaccard) with non-categorical kernels (Linear and RBF) and Random Forest (RF), including weighted and unweighted kernel variants.
• Weight derivation from RF: Derives weights from the decrease in node impurity as determined by Random Forest.
• Drug class coverage: Evaluated across 21 antiretroviral drugs spanning protease inhibitors (PI), integrase inhibitors (INI), nucleoside reverse transcriptase inhibitors (NRTI), and non-nucleoside reverse transcriptase inhibitors (NNRTI).
• Performance outcomes: Observed that the Jaccard kernel outperformed alternatives for 20 of the 21 drugs in the analysis.
• Protein-specific effects: Shows that weighting substantially improves prediction for reverse transcriptase-targeted drugs, with smaller gains for protease-targeted drugs related to differences in weight distribution measured by the Gini index.

Scientific Applications:

• HIV resistance prediction: Predicts drug resistance phenotypes from HIV sequence data to inform antiretroviral therapy selection.
• Method benchmarking for categorical sequence data: Provides a comparative framework for evaluating kernel methods and Random Forest on categorical, mixture-containing viral sequence datasets.

Methodology:

Applies weighted categorical kernel functions (Overlap and Jaccard), compares them with Linear and RBF kernels and Random Forest, derives residue weights from Random Forest decrease in node impurity, evaluates weighted and unweighted kernel variants across 21 antiretroviral drugs (PI, INI, NRTI, NNRTI), and assesses weight distribution using the Gini index while handling allele mixtures and residue-specific contributions.