HIVprotI

HIVprotI predicts inhibitory activity of compounds against HIV reverse transcriptase (RT), protease (PR), and integrase (IN) using computational models to support anti-HIV drug discovery.

Key Features:

• Support Vector Machine (SVM) regression models: SVM-based regression models are implemented to predict compound inhibition against PR, RT, and IN.
• QSAR-based molecular descriptors: Models use QSAR-derived molecular descriptors as input features.
• Training data from ChEMBL: Models are developed using experimentally validated IC50 and percent inhibition data sourced from the ChEMBL repository.
• Prediction outputs: Models predict IC50 values and percent inhibition for compounds against each protein target.
• Tenfold cross-validation performance: Tenfold cross-validation yielded Pearson correlation coefficients in the range 0.68–0.78 on IC50 and percent inhibition datasets.
• Independent dataset validation: Model performance is evaluated on independent datasets to assess generalizability.
• Chemical space mapping and applicability domain: Chemical space mapping and applicability domain analyses are performed to define model applicability and chemical coverage.
• Statistical validation tests: Additional statistical tests are applied to assess predictive validity.
• Descriptor importance analysis: The approach identifies chemical descriptors critical for predicting compound inhibition potential.

Scientific Applications:

• Virtual screening of inhibitors: Prioritize and rank compounds against RT, PR, and IN using predicted IC50 and percent inhibition.
• Lead design and optimization: Guide design and optimization of novel molecules targeting RT, PR, and IN based on QSAR descriptor insights.
• Prioritization for experimental testing: Select candidate compounds for in vitro assays according to predicted activity.
• Support for resistance-focused selection: Assist identification of compounds with predicted activity against targets relevant to drug-resistant HIV strains.

Methodology:

Support Vector Machine-based regression models trained on experimentally validated IC50 and percent inhibition data from ChEMBL using QSAR-based molecular descriptors; model assessment via tenfold cross-validation (Pearson r 0.68–0.78) and independent dataset testing; chemical space mapping, applicability domain analyses, and statistical tests applied; important chemical descriptors identified from model analyses.