SimBoost

SimBoost predicts continuous drug-target binding affinities using a read-across approach to capture the full spectrum of interactions for drug discovery.

Key Features:

• Continuous affinity prediction: Predicts continuous binding affinity values rather than binary interaction labels to represent true negatives through true positives.
• Read-across approach: Applies a read-across methodology to leverage similarity among compounds and targets for affinity prediction.
• Gradient boosting machines: Uses gradient boosting machines as the core predictive algorithm to model complex relationships in the data.
• SimBoostQuant prediction intervals: Provides prediction intervals via the SimBoostQuant variant to quantify uncertainty in predicted affinities.
• Applicability Domain metrics: Defines Applicability Domain metrics to assess the reliability and scope of individual predictions.
• Benchmark evaluation: Evaluated on two established drug-target interaction benchmark datasets and a newly proposed read-across cheminformatics dataset, reporting performance superior to previously reported models.

Scientific Applications:

• Drug-target affinity prediction: Generates continuous binding affinity estimates to support compound prioritization in drug discovery.
• Read-across cheminformatics: Enables read-across analyses to distinguish non-interacting and inactive drug-target pairs without binary thresholds.
• Uncertainty quantification: Uses prediction intervals and Applicability Domain metrics to inform confidence and reliability of affinity predictions.

Methodology:

Employs a read-across approach and gradient boosting machines for prediction; SimBoostQuant computes prediction intervals and defines Applicability Domain metrics; models were trained and evaluated on two established drug-target interaction benchmark datasets and a newly proposed read-across cheminformatics dataset.