Data mining a small molecule drug screening representative subset from NIH PubChem

Area of Science:

• Cheminformatics
• Computational Chemistry
• Drug Discovery

Background:

• PubChem is a large compound repository with over 10 million records, posing challenges for efficient drug screening.
• Facilitating information exchange and data sharing is crucial for the scientific community and NIH Roadmap Initiatives.

Purpose of the Study:

• To develop a data mining cheminformatics approach for constructing a representative and structure-diverse sublibrary from the large PubChem database.
• To create a manageable subset for efficient virtual screening and high-throughput screening (HTS).

Main Methods:

• Utilized whole-molecule chemistry-space matrix calculation with a cell-based partition algorithm to select a representative subset.
• Evaluated the subset using compound property analyses based on 1D and 2D molecular descriptors.
• Assessed self-similarity using 2D molecular fingerprints compared to the source library.

Main Results:

• Generated rePubChem, a representative subset of 540K compounds from 5.3 million.
• The subset maintains structural diversity and basic molecular properties with minimal similarity and redundancy.
• Demonstrated the subset's value for in silico virtual screening and in vitro HTS drug screening.

Conclusions:

• The rePubChem subset is a valuable, structure-diverse compound resource for drug screening.
• The established subset generation method is important for acquiring diverse compounds for efficient screening and library synthesis.
• This approach is applicable to data mining large compound databases for various scientific applications.