Plants synthesize antioxidant compounds as a defense mechanism against reactive oxygen species. Recently, plant-derived antioxidant compounds have attracted attention due to the increasing consumer awareness in the heath industry. However, traditional methods for measuring the antioxidant activity of these compounds are time-consuming and costly. Therefore, our study constructed a quantitative structure-activity relationship (QSAR) model that can predict antioxidant activity using graph convolutional networks (GCN) from plant structural data. The accuracy (Acc) of the model reached 0.6 and the loss reached 0.03. Although with lower accuracy than previously reported QSAR models, our model showed the possibility of predicting DPPH antioxidant activity in a wide range of plant compounds (phenolics, polyphenols, vitamins, etc.) based on their graph structure.

Introduction
Materials and Methods
    1. Data collection
    2. Molecular Descriptors
    3. QSAR Model
    4. Hyperparameter Optimization
    5. Implementation detail
Results and Discussion
Acknowledgement
References

• Hyeon Cheol Kim(Master’s course, Department of Environmental Materials Science, Institute of Agriculture and Life Science, Gyeongsang National University, Jinju, 52828, Korea)
• Si Young Ha(Research Professor, Department of Environmental Materials Science, Institute of Agriculture and Life Science, Gyeongsang National University, Jinju, 52828, Korea)
• Woo Seok Lim(Undergraduate Student, Department of Environmental Materials Science, Institute of Agriculture and Life Science, Gyeongsang National University, Jinju, 52828, Korea)
• Jae-Kyung Yang(Professor, Department of Environmental Materials Science, Institute of Agriculture and Life Science, Gyeongsang National University, Jinju, 52828, Korea) Corresponding author