Highly Accurate and Explainable Predictions of Small-Molecule Antioxidants for Eight In Vitro Assays Simultaneously...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in chemistry

Background:

• Small molecule antioxidants are crucial in various industries, protecting against oxidative damage.
• Traditional experimental discovery is time-consuming; computational approaches are underexplored.

Purpose of the Study:

• To develop a multitask self-supervised learning method for predicting antioxidant activity.
• To simultaneously assess small molecules across eight common in vitro antioxidant assays.

Main Methods:

• A functional-group-based alternating multitask self-supervised molecular representation learning (FG-BERT) model was proposed.
• The model was trained and evaluated on its ability to predict antioxidant activities.

Main Results:

• The FG-BERT model significantly outperformed baseline models in predictive performance.
• Achieved high average F1 (0.860), BA (0.880), ROC-AUC (0.954), and PRC-AUC (0.937) scores.
• Demonstrated reliable predictive capabilities and excellent interpretability, identifying key structural fragments.

Conclusions:

• The developed FG-BERT model provides a robust computational tool for antioxidant discovery.
• An online platform (AOP) was created to facilitate access to these prediction capabilities.
• This approach is expected to accelerate the identification of novel small-molecule antioxidants.