SimBoost: a read-across approach for predicting drug-target binding affinities using gradient boosting machines.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Accurate drug-target interaction prediction is crucial for drug discovery.
• Existing binary methods have limitations in differentiating inactive and unknown interactions.
• Binarization thresholds in current models affect the prediction of activity levels.

Purpose of the Study:

• To develop a computational method for predicting continuous drug-target binding affinities.
• To introduce a confidence assessment tool for predictions, defining Applicability Domain metrics.
• To improve upon existing models for drug-target interaction prediction.

Main Methods:

• Development of SimBoost for continuous binding affinity prediction.
• Introduction of SimBoostQuant for prediction intervals and confidence assessment.
• Evaluation on established and novel drug-target interaction benchmark datasets.

Main Results:

• SimBoost successfully predicts continuous binding affinities across the full interaction spectrum.
• SimBoostQuant provides explicit Applicability Domain metrics via prediction intervals.
• Both SimBoost and SimBoostQuant demonstrate superior performance compared to previous models on multiple datasets.

Conclusions:

• SimBoost and SimBoostQuant offer a more nuanced and reliable approach to drug-target interaction prediction.
• The continuous prediction and confidence assessment enhance the utility of computational methods in drug discovery.
• These methods advance the prediction of binding affinities and support read-across cheminformatics applications.