cheML.io: an online database of ML-generated molecules.

Area of Science:

• Computational Chemistry
• Machine Learning
• Cheminformatics

Background:

• * Machine learning (ML) algorithms are increasingly used for *de novo* molecule generation.
• * Existing compound databases may not cover the full chemical space for drug discovery.
• * There is a need for accessible, diverse virtual molecule libraries.

Purpose of the Study:

• * To develop an open-access database of virtual molecules generated by ML algorithms.
• * To provide a user-friendly web interface for accessing and exploring these molecules.
• * To enable on-demand molecule generation to meet specific user requirements.

Main Methods:

• * Utilized 10 different ML frameworks for *de novo* molecule generation.
• * Trained ML algorithms on samples from the ZINC database of commercially available compounds.
• * Merged generated molecules and their calculated properties into a unified database.
• * Developed a web interface with a drawing widget for molecule searching and retrieval.

Main Results:

• * Created a comprehensive database of ML-generated virtual molecules.
• * Integrated calculated properties for each generated molecule.
• * Implemented a web interface for intuitive data browsing and searching.
• * Enabled users to retrieve molecules with desired structures and properties.
• * Established a mechanism for on-demand molecule generation and database expansion.

Conclusions:

• * The developed database offers a valuable resource for drug discovery and chemical research.
• * The ML-driven approach expands chemical diversity beyond existing commercial libraries.
• * The interactive web interface facilitates efficient exploration and generation of novel molecular entities.