TL;DR: Using the “correct” version of the biological networks to represent the tissue- and functional- gene-gene relationship is critical for analyzing biological datasets. While many specialized, integrative biological graphs have been developed towards this goal, it is not always possible for a single graph to capture the full complexity of disease pathogenecity, such as in cancer.

Now, we developed a more powerful and data-driven approach, Explainable Multilayer Graph Neural Network (EMGNN), for predicting gene pathogenicity based on multiple input biological graphs. EMGNN maximizes the concordance of functional gene relationships with the unknown disease physiology by jointly modeling the multilayered graph structure. EMGNN achieves superior performance (average AUPR improvement 7.15%).

Michail’s paper is now accepted at Bioinformatics. You can read the advanced online version here:

Explainable Multilayer Graph Neural Network for Cancer Gene Prediction. Chatzianastasis, Michail, Vazirgiannis, Michalis, and Zhang, Zijun. Bioinformatics. Paper Link