PharmaBench: Enhancing ADMET benchmarks with large language models.

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Artificial intelligence in life sciences

Background:

• Accurate prediction of Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) properties is critical for successful drug development.
• Existing ADMET benchmark datasets are often small and lack representation of compounds relevant to current drug discovery projects, limiting their utility.
• These limitations hinder the development of robust predictive models for drug discovery.

Purpose of the Study:

• To develop a comprehensive and representative benchmark dataset for ADMET properties.
• To overcome the limitations of existing datasets in terms of size and compound representation.
• To facilitate the development of advanced AI models for drug discovery.

Main Methods:

• Utilized a multi-agent data mining system powered by Large Language Models to identify experimental conditions across 14,401 bioassays.
• Developed a data processing workflow to integrate and merge entries from diverse sources, resulting in 156,618 raw entries.
• Constructed PharmaBench, a benchmark set comprising eleven ADMET datasets with 52,482 entries.

Main Results:

• Successfully created PharmaBench, a large-scale, open-source benchmark dataset for ADMET properties.
• The dataset integrates data from multiple sources, enhancing its representativeness for drug discovery projects.
• PharmaBench contains 52,482 curated entries across eleven distinct ADMET datasets.

Conclusions:

• PharmaBench provides a valuable resource for developing and validating AI models in drug discovery.
• The dataset addresses the need for larger, more representative benchmark sets in ADMET research.
• This open-source resource is expected to accelerate the identification of drug candidates with favorable pharmacokinetic and toxicity profiles.