Multi-assay-based structure-activity relationship models: improving structure-activity relationship models by...

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Structure-activity relationship (SAR) models are crucial for optimizing drug leads in modern drug discovery.
• Current SAR models typically rely solely on activity data from the specific target of interest.

Purpose of the Study:

• To introduce and evaluate a novel class of multi-assay based SAR models.
• To leverage activity data from related targets to enhance SAR model performance.

Main Methods:

• Developed methods to identify related targets based on sequence or ligand structure.
• Employed machine learning techniques including semi-supervised learning, multi-task learning, and classifier ensembles.
• Utilized activity data from multiple assays to build SAR models.

Main Results:

• Multi-assay SAR models demonstrated significant improvements over standard models.
• Achieved an average 7.0-7.2% higher receiver-operating characteristic score on 117 PubChem protein targets.
• Outperformed chemogenomics-based approaches by 4.33% on targets within six protein families.

Conclusions:

• Multi-assay based SAR modeling represents a substantial advancement in drug discovery.
• These methods offer a more robust and accurate approach to predicting compound activity by integrating diverse biological data.