BEAR: A Novel Virtual Screening Method Based on Large-Scale Bioactivity Data.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Bioinformatics and systems biology

Background:

• Data-driven drug discovery leverages big data for efficient new drug development.
• Publicly available bioassay datasets offer extensive compound bioactivity profiles.
• Current virtual screening methods often rely on physicochemical features and structural information.

Purpose of the Study:

• To develop a novel in silico method, BEAR (Bioactive compound Enrichment by Assay Repositioning), for virtual screening of hit compounds.
• To utilize existing bioassay data for predicting compound activity against new targets through "assay repositioning".
• To enable scaffold hopping and identify structurally diverse drug candidates without requiring target or ligand physicochemical features.

Main Methods:

• Developed BEAR, an in silico method relying solely on bioactivity data for virtual screening.
• Performed large-scale cross-validation across over a thousand targets.
• Compared BEAR's performance against other machine learning models, including deep learning approaches.

Main Results:

• BEAR accurately predicted known ligands with a median area under the curve of 0.87 across diverse target classes.
• BEAR demonstrated robust performance, outperforming other machine learning models for ABC transporter family targets.
• Successfully predicted and validated nine novel P-gp and BCRP dual inhibitors using in vitro assays.

Conclusions:

• BEAR is a generally applicable and robust in silico method for drug discovery, leveraging bioactivity data effectively.
• The assay repositioning strategy allows for scaffold hopping and identification of structurally diverse compounds.
• BEAR successfully identified potent novel dual inhibitors for P-gp and BCRP, demonstrating its utility in identifying targeted therapeutics.