SeBioGraph: Semi-supervised Deep Learning for the Graph via Sustainable Knowledge Transfer.

Area of Science:

• Biomedical informatics
• Graph machine learning
• Sustainable manufacturing

Background:

• Semi-supervised deep learning is crucial for biomedical and advanced manufacturing graphs.
• Existing methods often fail on biomedical networks due to insufficient labeled data.
• Current graph neural networks are primarily evaluated on social/information networks, not biomedical ones.

Purpose of the Study:

• To propose a novel semi-supervised deep learning method, SeBioGraph, for biomedical graph analysis.
• To leverage sustainable knowledge transfer from auxiliary graphs to enhance target graph performance.
• To address limitations of traditional methods in scenarios with scarce labeled nodes.

Main Methods:

• SeBioGraph utilizes both node embedding and graph-specific prototype embedding.
• A transferable metric space is characterized using these embeddings.
• Prior knowledge from auxiliary graphs is incorporated to boost target graph performance.

Main Results:

• SeBioGraph achieves state-of-the-art results on node classification and link prediction tasks.
• The method demonstrates effectiveness in two-class node classification and three-class link prediction.
• Performance improvements are observed, particularly in data-scarce scenarios.

Conclusions:

• SeBioGraph offers a robust solution for semi-supervised learning on biomedical graphs.
• Sustainable knowledge transfer is a viable strategy for improving graph representation learning.
• The proposed method shows significant potential for applications in both biomedical and advanced manufacturing fields.