Related Experiment Video
Updated: Jun 3, 2025
07:22
Author Spotlight: Evaluating Biophysical Assays for Characterizing PROTACS Ternary Complexes
Published on: January 12, 2024
3.2K
Prediction of Proteolysis-Targeting Chimeras Retention Time Using XGBoost Model Incorporated with Chromatographic
Xinhao Qu1, Chen Jiang1,2, Mengyi Shan1
1School of Pharmaceutical Sciences, Zhejiang Chinese Medical University, Hangzhou 310053, People's Republic of China.
Journal of Chemical Information and Modeling
|January 9, 2025
Summary
Predicting retention times for Proteolysis-targeting chimeras (PROTACs) is now easier. A new machine learning model accurately forecasts PROTACs’ liquid chromatography retention times, aiding drug design and identification.
Area of Science:
- Medicinal Chemistry
- Analytical Chemistry
- Computational Chemistry
Background:
- Proteolysis-targeting chimeras (PROTACs) are heterobifunctional molecules for targeting previously undruggable proteins.
- Accurate structural identification and drug design of PROTACs are challenging due to their unique molecular structure.
- Liquid chromatography-mass spectrometry (LC-MS) is crucial for PROTACs annotation, but retention time (RT) prediction is difficult.
Purpose of the Study:
- To develop a predictive model for accurate retention time (RT) estimation of PROTACs in liquid chromatography (LC).
- To evaluate the performance of various machine learning algorithms for RT prediction in PROTACs.
- To identify key molecular and chromatographic features influencing PROTACs RT.
Main Methods:
- Compiled a comprehensive PROTAC-RT dataset from scientific literature.
- Evaluated machine learning models including XGBoost, RF, KNN, SVM, and FCNN using molecular fingerprints, descriptors, and chromatographic condition descriptors (CCs).
- Optimized the XGBoost model by screening feature combinations and performing hyperparameter tuning.
Main Results:
- An optimized XGBoost model incorporating moe206, Path, Charge descriptors, and CCs achieved high predictive accuracy (R² = 0.963 ± 0.023, RMSE = 0.896 ± 0.374).
- The model demonstrated strong performance on new chromatographic separation conditions and was validated experimentally.
- SHAP analysis highlighted the importance of CCs and molecular features like bond variability, van der Waals surface area, and atomic charge states.
Conclusions:
- The developed XGBoost model provides a fast and precise method for predicting PROTACs RT, significantly aiding their annotation.
- The findings underscore the utility of machine learning in accelerating drug discovery and development for complex molecules like PROTACs.
- Integration of molecular and chromatographic features is key for accurate RT prediction in PROTACs analysis.
