MTGNN: Multi-Task Graph Neural Network based few-shot learning for disease similarity measurement

Area of Science:

• Computational Biology
• Medical Informatics
• Pathology

Background:

• Identifying similar diseases aids in understanding complex disease mechanisms and improving diagnostics.
• Limited labeled data for similar disease pairs hinders the development of effective computational models.
• Disease relationships are often rooted in shared molecular origins or phenotypic similarities.

Purpose of the Study:

• To propose a novel Multi-Task Graph Neural Network (MTGNN) framework for measuring disease similarity using few-shot learning.
• To address the challenge of insufficient labeled similar disease pairs in training machine learning models.
• To leverage a multi-task learning strategy to enhance the accuracy of disease similarity prediction.

Main Methods:

• Developed a Multi-Task Graph Neural Network (MTGNN) framework incorporating few-shot learning.
• Implemented a multi-task optimization strategy, combining a disease similarity task with a link prediction task.
• Utilized high-dimensional disease embeddings derived from the dual-task learning process to quantify similarity.

Main Results:

• The MTGNN framework demonstrated strong performance in measuring disease similarity.
• The proposed method showed significant advantages over existing approaches, particularly with limited labeled training data.
• The multi-task learning strategy effectively compensated for the scarcity of labeled similar disease pairs.

Conclusions:

• The MTGNN framework offers a robust solution for disease similarity measurement in data-scarce scenarios.
• This approach enhances our ability to uncover pathogenic mechanisms and improve clinical outcomes for complex diseases.
• Few-shot learning combined with multi-task optimization is a promising direction for computational pathology research.