An interpretable geometric depth learning model for structure based protein binding site prediction

Today, I share an article published by Tel Aviv University in Nature Methods 2022 On the prediction of protein binding sites :《ScanNet: an interpretable geometric deep learning model for structure-based protein binding site prediction》. Predict the functional sites of proteins from their structures ( Such as small molecules 、 The binding site of a protein or antibody ), Can understand its function in the body . The author proposes an end-to-end 、 Interpretable geometry depth learning model , According to the spatial chemical arrangement between atoms and amino acids, the relationship between atoms and amino acids is established 3D characterization , And use them to predict the label of each amino acid （ Probability of belonging to binding site ）. The method is applied to protein - Protein binding sites PPBS forecast 、 Antibody binding site BCE It is predicted that good performance has been achieved on both tasks . Besides , Some of the model predictions SARS-CoV-2 Antigenic epitopes of spike proteins , Not only by known antigenic regions , In addition, previously undetected antigen regions were detected .

Article address ： https://www.nature.com/articles/s41592-022-01490-7.pdf