Predicting antioxidant activity of compounds based on chemical structure using machine learning methods.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Biochemistry and molecular biology

Background:

• Oxidative stress is linked to chronic diseases, necessitating effective antioxidant identification.
• Traditional antioxidant activity assays are time-consuming and resource-intensive.
• Predictive modeling offers a faster alternative for screening potential antioxidants.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) algorithms in predicting antioxidant activity.
• To identify the most accurate ML models for predicting antioxidant properties based on molecular structure.
• To assess the generalizability of developed ML models using external datasets.

Main Methods:

• Trained five ML algorithms: Support Vector Machine (SVM), Logistic Regression (LR), XGBoost, Random Forest (RF), and Deep Neural Network (DNN).
• Utilized a dataset of over 1,900 compounds with experimentally determined antioxidant activity for training and validation.
• Performed external validation using natural product data from the BATMAN database.

Main Results:

• Random Forest (RF) and Support Vector Machine (SVM) models demonstrated superior performance with accuracy exceeding 0.9.
• These models effectively differentiated between active and inactive compounds, even those with high structural similarity.
• External validation confirmed the robust generalizability of the RF and SVM models.

Conclusions:

• Machine learning models, particularly RF and SVM, are powerful tools for predicting antioxidant activity.
• ML-based approaches can significantly accelerate the identification of novel antioxidant candidates.
• This methodology holds potential for streamlining the development of therapeutic interventions targeting oxidative stress.