Estimation of the optimal data base size for structure-activity analyses: the Salmonella mutagenicity data base

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Structure-Activity Relationship (SAR) models are crucial for drug discovery and development.
• The size and quality of the training dataset significantly influence SAR model performance.
• Understanding the impact of data size is essential for building robust predictive models.

Purpose of the Study:

• To investigate the effect of database size on the predictivity and informational content of SAR models.
• To determine the optimal database size for deriving reliable SAR models.
• To analyze the relationship between database size and structural overlap in SAR models.

Main Methods:

• Systematic evaluation of SAR models derived from datasets of varying sizes.
• Assessment of model predictivity using metrics such as sensitivity, specificity, and concordance (OCP).
• Quantification of informational content and structural overlap as a function of database size.

Main Results:

• Predictivity indices (sensitivity, specificity, OCP) increased with database size up to 300-400 chemicals, then plateaued.
• Informational content of SAR models generally increased with database size.
• The rate of increase in informational content diminished for databases exceeding 400 chemicals, indicating a point of diminishing returns.

Conclusions:

• Database size is a critical factor influencing SAR model performance and informational content.
• An optimal range of 300-400 chemicals exists for maximizing predictivity in SAR models.
• Careful consideration of database size is necessary to balance model accuracy and computational efficiency in cheminformatics applications.