What question did this study set out to answer?

The aim is to develop a pharmacokinetic property prediction model for EGFR-targeting PROTACs using machine learning.

February 19, 2026Open Access

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

Puntos clave

The aim is to develop a pharmacokinetic property prediction model for EGFR-targeting PROTACs using machine learning.
Developed a pharmacokinetic prediction model called EGFR-PROPK.
Combined traditional machine learning with molecular fingerprints.
Conducted in-vivo experiments on 100 EGFR-targeting PROTAC molecules.
Focused on key pharmacokinetic parameters: clearance, half-life, and volume of distribution.
Traditional models perform poorly for PROTACs.
Training on PROTAC-specific data improved prediction accuracy.
Achieved correlation coefficients of 0.78 for half-life, 0.75 for clearance, and 0.52 for volume of distribution.

Resumen

PROteolysis TArgeting Chimeras (PROTACs) are bifunctional molecules that offer a novel approach to targeted protein degradation, showing particular promise for previously 'undruggable' targets. Despite their therapeutic potential, significant challenges remain in optimizing the pharmacokinetic (PK) properties of PROTACs, particularly in terms of their ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics. In this study, we drew on the experience of previous work and combined traditional machine learning with multiple molecular fingerprints to propose a PROTAC pharmacokinetic property prediction model EGFR-PROPK. We conducted in-vivo experiments on 100 EGFR-targeting PROTAC molecules, focusing on clearance (CL), half-life (T1/2), and apparent volume of distribution (Vss), which are the most intuitive and commonly used macro parameters to describe the PK properties of a drug since they describe the absorption-distribution-elimination process of drugs in the body from different perspectives. Our findings reveal that traditional models trained on small molecules perform poorly on PROTACs. However, training the models on PROTAC-specific data significantly improved prediction accuracy, achieving correlation coefficients of 0.78, 0.75, and 0.52 between the predicted and observed values for T1/2, CL and Vss, respectively, highlighting the need for tailored approaches in PK evaluation for these unique molecules. These insights are critical for advancing the design and development of PROTAC-based therapies.

Me gusta

Guardar

Ver artículo completo