QSAR modeling of imbalanced high-throughput screening data in PubChem

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Quantitative Structure-Activity Relationship (QSAR) modeling

Background:

• PubChem contains numerous structures with high-throughput screening (HTS) assay data.
• HTS assay data are characteristically imbalanced, with a predominance of inactive compounds over active ones.
• This imbalance poses challenges for building reliable predictive models.

Purpose of the Study:

• To develop and test strategies for efficiently building robust QSAR models from imbalanced HTS data.
• To compare different descriptor types and modeling approaches for QSAR development.
• To integrate predictive QSAR models into publicly available web services.

Main Methods:

• Utilized imbalanced PubChem HTS assay datasets.
• Generated QSAR models using Quantitative Neighborhoods of Atoms (QNA) and biological descriptors within the GUSAR program.
• Evaluated model performance using external test and validation sets.

Main Results:

• Successfully developed and tested strategies for building QSAR models from imbalanced HTS data.
• Identified effective descriptor types and modeling approaches for robust QSAR generation.
• Demonstrated the potential for integrating predictive models into web services.

Conclusions:

• Robust QSAR models can be effectively built from imbalanced HTS data using appropriate strategies and descriptors.
• The developed QSAR modeling approaches show promise for enhancing drug discovery efforts.
• Integration into web services facilitates broader accessibility and application of predictive modeling tools.