We recently published a new paper in the journal Cell Reports, in which we present the development and application of graph convolutional neural networks (GCNs) to the analysis of complex carbohydrates or glycans. As GCNs can easily work with the diversity and nonlinearity of glycans (they are frequently branched), our new model outperforms previous model architectures in all prediction tasks. We also demonstrate new applications, such as comparing glycomes between species, based on learned similarities, to detect glycome-phenome relations. Lastly, we demonstrate that we can use glycan-focused GCNs for predicting and analyzing virus-glycan interactions to detect new viral glycan receptors relevant for infection.
If you want to learn more, we have a more in-depth article about our work that is published in Towards Data Science. Additionally, here are some easily digestible press releases about this paper that discuss some of the implications our work could have. Also, here is the code for the model discussed here; all data can be found in the supplementary tables of our open-access paper. Feel free to reach out if you are interested in this area, there is still much to be done in glycan-focused machine learning and data science!