HDAC3i-Finder

HDAC3i-Finder identifies potential histone deacetylase 3 (HDAC3) inhibitors using machine-learning models to support virtual screening in drug discovery.

Key Features:

• Machine Learning Model: The XGBoost_Morgan2 model uses XGBoost with Morgan2 fingerprints (1024 bits) as molecular features for HDAC3 inhibitor prediction.
• Data Compilation and Feature Calculation: A dataset of 1098 compounds from ChEMBL was compiled and for each compound Mordred two-dimensional descriptors, MACCS keys (166 bits), and Morgan2 fingerprints were calculated.
• Model Training and Optimization: Five classifiers—k-Nearest Neighbour (KNN), Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Deep Neural Network (DNN)—were trained on each feature set, yielding 15 models.
• Performance Evaluation: XGBoost_Morgan2 was benchmarked on the MUBD-HDAC3 set and achieved an early ROC enrichment of ROCE0.5% = 41.02.
• Retrospective Screening: Retrospective screening of an annotated PubChem chemical library with XGBoost_Morgan2 identified eight novel-scaffold HDAC3 inhibitors by assaying 1% of compounds.

Scientific Applications:

• Virtual screening: Prioritizing candidate molecules for experimental testing as HDAC3 inhibitors in drug discovery campaigns.
• Novel-scaffold identification: Discovering structurally novel HDAC3 inhibitors through retrospective and prospective library screening.
• Disease-focused research: Supporting studies of HDAC3-targeted therapeutics in cancer, chronic inflammation, neurodegenerative disorders, and diabetes.

Methodology:

Compilation of 1098 ChEMBL compounds; calculation of Mordred 2D descriptors, MACCS keys (166 bits), and Morgan2 fingerprints (1024 bits); training of KNN, SVM, RF, XGBoost, and DNN classifiers on each feature set; selection and benchmarking of XGBoost_Morgan2 on MUBD-HDAC3 with ROCE0.5% metric; retrospective screening of a PubChem annotated library.