EGFRisopred: a machine learning-based classification model for identifying isoform-specific inhibitors against EGFR and HER2

Area of Science:

• Oncology
• Computational Chemistry
• Drug Discovery

Background:

• The Epidermal Growth Factor Receptor (EGFR) kinase pathway is crucial in human cancers.
• EGFR and Human Epidermal growth factor Receptor 2 (HER2) are key targets for lung and breast cancer therapies.
• Developing isoform-specific inhibitors for EGFR and HER2 presents a significant challenge in cancer drug discovery.

Purpose of the Study:

• To create a knowledge-based computational classification model for predicting molecule specificity towards EGFR or HER2.
• To identify prevalent molecular scaffolds and functional groups associated with EGFR- and HER2-specific inhibitors.
• To develop a user-friendly application for predicting anticancer agent specificity.

Main Methods:

• Collected a dataset of 519 compounds with inhibitory activity against EGFR and HER2.
• Developed 72 classification models using nine fingerprint types and four classifiers (IBK, NB, SMO, RF).
• Analyzed scaffolds and functional groups; evaluated model accuracy using a decoy dataset.

Main Results:

• Models using Random Forest and IBK classifiers showed superior performance for EGFR- and HER2-specific datasets, respectively.
• Identified prevalent core structures and fragments within EGFR- and HER2-specific compound datasets.
• Developed the EGFRisopred application integrating best-performing models for predicting compound specificity.

Conclusions:

• The developed computational models and the EGFRisopred tool can aid researchers in identifying novel EGFR/HER2-specific inhibitors.
• This approach facilitates the discovery of targeted anticancer agents for lung and breast cancer.
• The free utility is expected to accelerate the identification of new inhibitors for these critical cancer targets.