Development of QSAR Models and Web Applications for Predicting hDHFR Inhibitor Bioactivity Using Machine Learning.

Area of Science:

• Biochemistry and Medicinal Chemistry
• Computational Chemistry and Cheminformatics

Background:

• Human dihydrofolate reductase (hDHFR) is a key enzyme in folate metabolism, essential for DNA synthesis and cell proliferation.
• hDHFR is a validated therapeutic target for anticancer, antimicrobial, and antiprotozoal treatments.
• Emerging resistance to current hDHFR inhibitors necessitates novel drug development.

Purpose of the Study:

• To develop predictive machine learning models for hDHFR bioactivity using cheminformatics approaches.
• To identify key molecular features associated with hDHFR inhibition.
• To create a user-friendly web application for predicting hDHFR bioactivity of new compounds.

Main Methods:

• Utilized three molecular descriptor types (PubChem, Substructure, MACCS fingerprints) to represent compound structures.
• Employed a random forest algorithm with hyperparameter tuning for model building.
• Applied Recursive Feature Elimination (RFE) for feature selection and Principal Component Analysis (PCA) for dimensionality reduction and outlier removal.

Main Results:

• Achieved high predictive performance with R-squared values of 0.9849-0.9934 (training) and 0.9381-0.9591 (test sets).
• Validated models using external test sets and domain applicability analysis.
• Identified influential molecular features contributing to hDHFR inhibition through feature importance analysis.

Conclusions:

• The developed cheminformatics models accurately predict hDHFR bioactivity.
• These models can accelerate the discovery of novel hDHFR inhibitors to overcome drug resistance.
• An accessible web application is available for researchers to screen potential drug candidates.