Essential to all living organisms, glycans (sugars) bind to a diverse array of glycan binding proteins.  We can measure the strength of this binding experimentally, but this requires purifying each glycan of interest and each protein of interest, and then conducting the experiments to measure the binding.  We believe that ML can extrapolate from already published values, and predict glycan-protein binding in new and unmeasured cases.

We hope to see a model that takes a glycan (or similar) chemical structure and an amino acid sequence of a protein, and produces accurate predictions for the binding strength between the pair.

We have provided training data derived from the Consortium for Functional Glycomics (https://www.functionalglycomics.org ) and BindingDB (https://www.bindingdb.org ).

We have run a similar competition before, but this time we have provided more data and we hope that the Kaggle community can find better models this time around.