Characterisation of data resources for in silico modelling: benchmark datasets for ADME properties

Area of Science:

• Computational chemistry
• Pharmacology
• Drug discovery

Background:

• High-quality data are crucial for developing accurate in silico Absorption, Distribution, Metabolism, and Excretion (ADME) models.
• The cost of traditional in vivo and in vitro screening necessitates the development of computational models.
• Dataset characteristics like size, format, and chemical identifiers impact modelability.

Purpose of the Study:

• To review the utility of publicly available ADME datasets for predictive model development.
• To assess the modelability of selected ADME datasets based on specific criteria.
• To provide recommendations for dataset suitability assessment and data publication.

Main Methods:

• Collated over 140 ADME datasets from public resources.
• Assessed the modelability of 31 selected datasets using study-derived criteria.
• Analyzed dataset characteristics including size, format, and chemical identifiers.

Main Results:

• Publicly available ADME datasets vary significantly in content and presentation.
• Adequate dataset size, user-friendly format, and appropriate chemical identifiers (e.g., CAS, SMILES, InChIKey) are essential for modeling.
• Specific criteria were developed to evaluate dataset suitability for predictive modeling.

Conclusions:

• Researchers must carefully assess publicly available ADME datasets for quality and suitability.
• Standardized data formats and comprehensive identifiers enhance the development of reliable in silico ADME models.
• Recommendations are provided for both assessing and publishing ADME datasets to facilitate model development.