A machine learning-based pharmacokinetics predictor (EGFR-PROPK) for EGFR-targeting PROTACs.

Area of Science:

• Drug Discovery
• Pharmacology
• Biotechnology

Background:

• PROteolysis TArgeting Chimeras (PROTACs) offer a novel strategy for targeted protein degradation, addressing previously undruggable targets.
• Optimizing pharmacokinetic (PK) properties, including ADMET, remains a significant hurdle for PROTAC development.
• Traditional machine learning models often struggle with the unique characteristics of PROTACs compared to small molecules.

Purpose of the Study:

• To develop and validate a PROTAC-specific pharmacokinetic property prediction model.
• To evaluate the predictive performance of traditional machine learning models versus PROTAC-tailored models.
• To improve the understanding of PK parameters (CL, T1/2, Vss) for EGFR-targeting PROTACs.

Main Methods:

• Combined traditional machine learning with multiple molecular fingerprints.
• Developed the EGFR-PROPK model for PROTAC pharmacokinetic property prediction.
• Conducted in-vivo experiments on 100 EGFR-targeting PROTAC molecules to assess CL, T1/2, and Vss.

Main Results:

• Traditional models trained on small molecules showed poor performance when applied to PROTACs.
• Training models on PROTAC-specific data significantly enhanced prediction accuracy.
• Achieved correlation coefficients of 0.78 for T1/2, 0.75 for CL, and 0.52 for Vss between predicted and observed values.

Conclusions:

• PROTAC pharmacokinetic evaluation requires tailored approaches distinct from small molecule drug development.
• The EGFR-PROPK model demonstrates the effectiveness of PROTAC-specific data in improving PK prediction.
• These findings are crucial for advancing the rational design and development of PROTAC-based therapeutics.