Protein's cavities Bestiario: an atlas of proteins cavities

ANA JULIA VELEZ RUEDA; BULGARELLI, FRANCO LEONARDO; GUSTAVO PARISI; LEANDRO BUGNON

Conferencia; KHIPU Latin America Meeting in Artificial Intelligence; 2023

On their surface proteins are shaped into numerous cavities and protrusions that create unique microenvironments for ligand binding, catalysis or any other biologically relevant. Despite their biological and biotechnological relevance and potential impact on various research areas in medicine, drug design, and evolutionary biology, there is still no exhaustive and comprehensive research about protein cavities that combines sequential, structural, and evolutionary perspectives.We propose a representative feature set to enable machine learning tasks such as clustering and classifying the different types of protein's cavities in functional categories. For such purposes, we combined classical bioinformatic techniques for molecular characterization with unsupervised learning techniques. We created and curated a dataset with all the proteins with known structures and their features, currently available in CaviDB 1. In this dataset, we have computationally predicted the cavities present in each of the proteins and characterized them structurally and sequentially on a set of 47 features.  We explored the data  using non-linear projections with UMAP and Self-Organized Maps. Using biological information such as volume, drug content, and polarity, we were able to cluster representative cavities and use them to reconstruct protein families.