EGFR^AP: a predictive machine learning model for assessing small molecule activity against the epidermal growth factor...

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Epidermal growth factor receptor (EGFR) is crucial for cell growth, but its mutations drive tumorigenesis.
• Targeting EGFR is a key strategy in cancer therapy.
• Developing novel EGFR inhibitors requires efficient predictive tools.

Purpose of the Study:

• To develop a machine learning tool, EGFR^AP, for predicting the biological activity (pIC₅₀) of novel molecules against EGFR.
• To create a robust quantitative structure-activity relationship (QSAR) model for EGFR inhibitor discovery.

Main Methods:

• Utilized a large dataset of existing EGFR inhibitors to train machine learning algorithms.
• Developed a QSAR model using the extra trees regressor (ET) algorithm.
• Validated the model's performance using training and test datasets, and 10-fold cross-validation.
• Performed structure-based drug design experiments to confirm predictions.

Main Results:

• The ET model achieved high accuracy on the training set (R²=0.99, RMSE=0.07, MAE=0.009).
• Validation on the test set showed satisfactory performance (R²=0.67, RMSE=0.89, MAE=0.61).
• Cross-validation and drug design experiments confirmed the model's robustness and predictive power.

Conclusions:

• The EGFR^AP tool demonstrates significant potential for identifying novel EGFR inhibitors.
• This computational approach can accelerate the drug discovery process for medicinal chemists.
• EGFR^AP is a valuable asset for developing targeted cancer therapies.