HDAC3i-Finder: A Machine Learning-based Computational Tool to Screen for HDAC3 Inhibitors

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Histone deacetylase 3 (HDAC3) is a promising therapeutic target for various human diseases, including cancer, chronic inflammation, neurodegenerative disorders, and diabetes.
• Quantitative Structure-Activity Relationship (QSAR) modeling using machine learning (ML) is a powerful cheminformatics approach for drug discovery.
• Previous QSAR studies have not specifically focused on HDAC3 inhibitors.

Purpose of the Study:

• To develop and optimize machine learning models for predicting HDAC3 inhibitors.
• To identify novel chemical scaffolds with potential HDAC3 inhibitory activity.
• To create a user-friendly tool for facilitating the screening of HDAC3 inhibitors.

Main Methods:

• Compiled a dataset of 1098 compounds with known HDAC3 activity from the ChEMBL database.
• Calculated three sets of molecular features: Mordred descriptors, MACCS keys, and Morgan2 fingerprints.
• Trained and compared five ML classifiers (KNN, SVM, RF, XGBoost, DNN) on each feature set, generating 15 models.

Main Results:

• The XGBoost model utilizing Morgan2 fingerprints (XGBoost_morgan2) demonstrated superior performance.
• This best model achieved a high ROC enrichment of 41.02% on the MUBD-HDAC3 benchmark set.
• Retrospective screening identified 8 novel-scaffold HDAC3 inhibitors from a PubChem library by analyzing only 1% of compounds.

Conclusions:

• The developed XGBoost_morgan2 model is effective for identifying potential HDAC3 inhibitors.
• The HDAC3i-Finder Python GUI application provides accessible tool for the scientific community for prospective screening.
• This approach accelerates the discovery of novel HDAC3 inhibitors for therapeutic applications.