Machine learning-based chemical binding similarity using evolutionary relationships of target genes

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Chemical similarity searching is vital for identifying molecules with similar shapes to known active compounds.
• Traditional methods often fail to retrieve critical local molecular features essential for target binding.
• A need exists for improved methods that capture functional activity related to target interactions.

Purpose of the Study:

• To develop a novel machine learning-based chemical binding similarity score.
• To leverage evolutionary relationships of binding targets to enhance similarity searching.
• To improve the retrieval of molecules with functional activity for target binding.

Main Methods:

• Integrated comprehensive and heterogeneous multiple target-binding chemical data into a paired format.
• Employed multiple classification similarity-learning models incorporating target evolutionary information.
• Defined chemical similarity as the probability of compounds binding to identical targets.

Main Results:

• Encoding target evolutionary information significantly expanded chemical-target interaction data.
• The ensemble evolutionary chemical binding similarity (ensECBS) model demonstrated improved performance.
• The model effectively identified hidden chemical relationships and enhanced similarity searching.

Conclusions:

• The developed ensECBS method offers a novel approach to chemical similarity searching.
• Utilizing evolutionarily conserved target binding information enhances the discovery of functional molecular relationships.
• This approach can serve as a valuable tool in drug discovery and chemical biology.