Perspectives on the use of machine learning for ADME prediction at AstraZeneca

Area of Science:

• Drug discovery and development
• Pharmacokinetics and drug metabolism
• Computational chemistry and cheminformatics

Background:

• Understanding a drug's pharmacokinetic (PK) profile is crucial for determining dosage, administration frequency, and potential adverse reactions.
• Early prediction of PK properties in the drug discovery process is essential for improving efficiency and reducing late-stage failures.
• Current methods often require experimental data, highlighting the need for predictive models prior to molecule synthesis.

Purpose of the Study:

• To describe AstraZeneca's approaches for enhancing the prediction of preclinical and human pharmacokinetic profiles of novel molecules.
• To leverage machine learning (ML) and artificial intelligence (AI) for improved PK prediction.
• To integrate chemical structure-based methods with experimental data for more accurate in vivo pharmacokinetic predictions.

Main Methods:

• Utilizing machine learning and artificial intelligence algorithms to analyze chemical structures and experimental data.
• Combining in silico (structure-based) approaches with in vitro experimental properties.
• Developing predictive models for both preclinical and human pharmacokinetic parameters.

Main Results:

• Demonstrated improved prediction of in vivo pharmacokinetics by integrating chemical structure and experimental data.
• Extended predictive capabilities to molecules outside the traditional Lipinski's rule-of-five space.
• Showcased the potential of combined in vitro and in vivo predictive models for anticipating human outcomes.

Conclusions:

• Machine learning and AI significantly enhance the prediction of drug pharmacokinetic profiles.
• Integrating diverse data sources (chemical structure, in vitro, in vivo) leads to more robust PK predictions.
• These predictive capabilities can optimize chemical design and improve the efficiency of drug discovery.