DeepMolecules: a web server for predicting enzyme and transporter-small molecule interactions

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Understanding protein-small molecule interactions is crucial for various biological processes and drug development.
• Accurate prediction of enzyme kinetics (kcat, KM) and substrate identification is essential for optimizing biocatalysts and drug efficacy.

Purpose of the Study:

• To develop and provide an accessible web server, DeepMolecules, for predicting protein-small molecule interactions and enzyme kinetic parameters.
• To integrate multiple state-of-the-art predictive models into a user-friendly platform.

Main Methods:

• Utilized deep learning for numerical representations of proteins and small molecules.
• Employed gradient-boosted decision tree models for interaction and kinetic predictions.
• Developed a web interface supporting various input formats (SMILES, InChI, KEGG ID) and submission types (single, batch).

Main Results:

• Achieved high predictive performance for enzyme substrate identification, transporter substrate identification, enzyme turnover number (kcat), and Michaelis constant (KM).
• Integrated experimental data for a comprehensive view of protein-small molecule relationships.

Conclusions:

• DeepMolecules offers a powerful, freely accessible tool for researchers in metabolic engineering, drug discovery, and biocatalysis.
• The server facilitates the identification of potential substrates and the quantification of their catalytic properties, accelerating research and development.