Significantly improved HIV inhibitor efficacy prediction employing proteochemometric models generated from...

Area of Science:

• Computational chemistry
• Virology
• Pharmacology

Background:

• HIV treatment relies on combination antiretroviral therapy.
• Individualized treatment selection is crucial due to viral genetic diversity.
• Current models lack consideration of drug chemical similarity.

Purpose of the Study:

• To evaluate the added value of incorporating chemical similarity into HIV drug resistance prediction models.
• To develop and validate proteochemometric (PCM) models for predicting treatment outcomes.

Main Methods:

• Applied PCM models combining chemical and protein target properties to a large clinical dataset (300,000+ data points).
• Utilized genotypic and phenotypic information for HIV Reverse Transcriptase and Protease inhibitors.
• Compared PCM model performance against sequence-data-only models.

Main Results:

• PCM models achieved prediction error below 0.5 Log Fold Change.
• PCM models outperformed sequence-data models in resistance classification and Log Fold Change prediction (0.76 vs. 0.91 log units).
• Successfully identified known and novel resistance-conferring mutations and achieved 84% prospective prediction accuracy.

Conclusions:

• Proteochemometric models accurately predict HIV phenotypic resistance from genotypic data, including novel mutations and mixtures.
• The inclusion of chemical similarity enhances prediction accuracy and reliability.
• Developed models offer an applicability domain to guide clinical decision-making.