Daniel Bojar presents a deep-learning oracle to predict protein–glycan interactions
Glycan-binding proteins or lectins are crucial for understanding the interactions in biological systems. But because of the sheer breadth and depth of specificity of lectins studying them is largely a piecemeal work. The new AI-platform, called LectinOracle, uses a curated data set of 564,647 unique protein–glycan interactions to predict the specificities of a wide range of lectins. In the paper it is shown that LectinOracle predictions generalize to new glycans and lectins, with qualitative and quantitative agreement with experimental data. It is also shown that the platform can further be used to improve lectin classification, accelerate lectin directed evolution, investigate interactions of the microbiome, and predict epidemiological outcomes in the context of influenza virus. The researchers are sure that LectinOracle will advance both the study of lectins and their role in (glyco-)biology.
Advanced Science has highlighted the paper in the editorial of the January issue and chose the artwork as a frontispiece cover page.