Machine learning based approaches for structure activity relationship analysis of heparanase inhibitors

Area of Science:

• Biochemistry and Medicinal Chemistry
• Computational Drug Discovery
• Machine Learning in Pharmacology

Background:

• Human Heparanase (HPSE) is crucial in extracellular matrix remodeling and growth factor release.
• HPSE overexpression correlates with tumor growth, angiogenesis, metastasis, and inflammation, making it a key therapeutic target.
• Developing effective HPSE inhibitors is vital for treating oncology and inflammatory conditions.

Purpose of the Study:

• To create and validate a computational workflow for predicting the activity class of potential HPSE inhibitors.
• To leverage curated bioactivity data from the ChEMBL database for model development.
• To establish a tool for aiding virtual screening and hit prioritization in HPSE-targeted drug discovery.

Main Methods:

• Extracted and meticulously curated HPSE inhibitor bioactivity data from ChEMBL.
• Employed diverse molecular representations: 2D physicochemical descriptors, Morgan fingerprints, and 3D descriptors.
• Utilized machine learning classifiers with feature engineering, PCA, and SMOTE for robust prediction model development.

Main Results:

• The best-performing Random Forest model achieved ~80% accuracy and 78.5% balanced accuracy on a held-out test set.
• The validated computational workflow successfully predicted activity classes for novel compounds.
• The developed pipeline offers a reliable method for classifying HPSE inhibitors.

Conclusions:

• A robust computational workflow for predicting HPSE inhibitor activity classes has been successfully developed and validated.
• This tool can significantly aid in the virtual screening and prioritization of drug candidates targeting HPSE.
• The findings support the use of computational approaches in accelerating the discovery of therapeutics for cancer and inflammatory diseases.