ChEMBL-Likeness Score and Database GDBChEMBL.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• The GDB17 database contains 166.4 billion molecules, but many are too complex for practical synthesis.
• Existing subsets like FDB17 and GDBMedChem have limitations in synthetic accessibility or scope.

Purpose of the Study:

• To create a curated subset of GDB17 with enhanced drug-likeness and synthetic accessibility.
• To provide a valuable resource for medicinal chemistry and drug discovery efforts.

Main Methods:

• Developed a ChEMBL-likeness score (CLscore) based on substructure frequencies in the ChEMBL database.
• Selected 10 million molecules from GDB17 using CLscore, uniform sampling for molecular properties, and chemical stability rules.
• Ensured broad chemical space coverage and continuous distribution of molecular properties.

Main Results:

• GDBChEMBL comprises 10 million molecules with high synthetic accessibility and potential bioactivity.
• The subset retains the broad chemical space coverage characteristic of the parent GDB17 database.
• GDBChEMBL offers improved properties compared to previous GDB17 subsets.

Conclusions:

• GDBChEMBL represents a significant advancement in generating synthetically accessible and drug-like chemical spaces.
• This resource facilitates exploration of novel chemical matter for drug discovery.
• The database and interactive map are publicly available for research use.