X2Graph for Cancer Subtyping Prediction on Biological Tabular Data.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Deep learning excels with large datasets but struggles with scarce medical tabular data.
• Existing methods have limitations in analyzing small, complex biological datasets.

Purpose of the Study:

• To introduce X2Graph, a novel deep learning method for small biological tabular datasets.
• To improve cancer subtyping using limited data by leveraging external biological knowledge.

Main Methods:

• X2Graph converts tabular data samples into graph structures.
• It incorporates external knowledge, like gene interactions, into the graph representation.
• Standard message-passing algorithms are applied to these graphs for modeling.

Main Results:

• X2Graph demonstrated superior performance on three cancer subtyping datasets.
• Outperformed existing tree-based and deep learning methods in small dataset scenarios.

Conclusions:

• X2Graph offers a powerful deep learning solution for analyzing scarce tabular data in medicine.
• This method advances deep learning applications in cancer diagnosis with limited data.