StackNAFLD: An Accurate Stacking Ensemble Learning Targeting NAFLD Treatment.

Area of Science:

• Biomedical Informatics
• Computational Chemistry
• Drug Discovery

Background:

• Nonalcoholic fatty liver disease (NAFLD) presents complex pathophysiological mechanisms, making treatment challenging.
• Predicting molecules to inhibit NAFLD progression requires advanced computational approaches.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based stacking ensemble model for predicting NAFLD inhibitory agents.
• To identify key molecular features associated with NAFLD inhibition.

Main Methods:

• Collected 75 agents from preclinical studies, classifying them as inducers or inhibitors.
• Computed 12 molecular fingerprints and trained three baseline ML models.
• Developed a stacking ensemble model trained on baseline predictions and validated using 5-fold cross-validation and LOOCV.

Main Results:

• The stacking ensemble model demonstrated superior performance over baseline models in predicting NAFLD inhibitory activity.
• The model's robustness and applicability domain were validated, ensuring trustworthy predictions.
• Key molecular features, including carboxylic, alkene, and aromatic rings, were identified as influential.

Conclusions:

• Stacking ensemble learning offers an effective method for improving molecular property prediction in NAFLD research.
• The developed model and associated software are available on GitHub to support the drug discovery pipeline.