Predicting anticancer hyperfoods with graph convolutional networks.

Area of Science:

• Nutritional Science
• Computational Biology
• Bioinformatics

Background:

• Nutritional science is exploring bioactive food molecules for disease treatment by comparing them to FDA-approved drugs.
• Previous methods used separate unsupervised and supervised algorithms, leading to suboptimal drug representations and extensive computational resources for optimization.
• Representing drugs as graphs of their targets within the human interactome offers a new approach.

Purpose of the Study:

• To develop an end-to-end graph convolutional model for predicting anticancer molecules in food.
• To overcome the limitations of disjoint feature generation and classification tasks in previous methodologies.
• To optimize the prediction of anticancer therapeutics by learning improved drug representations.

Main Methods:

• Formulated drug prediction as a graph classification task, representing drugs as graphs with targets as node features.
• Employed an end-to-end graph neural network (GNN) model for direct graph operation and representation learning.
• Utilized the GNN to optimize performance in predicting anticancer therapeutics.

Main Results:

• The proposed GNN model achieved an F1 score of 67.99%±2.52% and an Area Under the Precision-Recall Curve (AUPR) of 73.91%±3.49% in anticancer therapeutic prediction.
• The model demonstrated the ability to capture biological pathway knowledge for predicting anticancer molecules.
• Outperformed the baseline approach in the prediction task.

Conclusions:

• Introduced an interpretable end-to-end graph convolutional model for identifying cancer-fighting food molecules.
• The model's superior performance advances the field of personalized nutrition for cancer prevention and therapeutics.
• Paves the way for developing targeted nutrition strategies based on molecular insights.