The identification and characterization of the structural sites which contribute to protein function are crucial for understanding biological mechanisms, evaluating disease risk, and developing targeted therapies. However, the quantity of known protein structures is rapidly outpacing our ability to functionally annotate them. Existing methods for function prediction either do not operate on local sites, suffer from high false positive or false negative rates, or require large site-specific training datasets, necessitating the development of new computational methods for annotating functional sites at scale. We present COLLAPSE (Compressed Latents Learned from Aligned Protein Structural Environments), a framework for learning deep representations of protein sites. COLLAPSE operates directly on the 3D positions of atoms surrounding a site and uses evolutionary relationships between homologous proteins as a self-supervision signal, enabling learned embeddings to implicitly capture structure-function relationships within each site. Our representations generalize across disparate tasks in a transfer learning context, achieving state-of-the-art performance on standardized benchmarks (protein-protein interactions and mutation stability) and on the prediction of functional sites from the Prosite database. We use COLLAPSE to search for similar sites across large protein datasets and to annotate proteins based on a database of known functional sites. These methods demonstrate that COLLAPSE is computationally efficient, tunable, and interpretable, providing a general-purpose platform for computational protein analysis.

A rich diversity of transmembrane G protein-coupled receptors (GPCRs) are used by eukaryotes to sense physical and chemical signals. In humans alone, 800 GPCRs comprise the largest and most therapeutically targeted receptor class. Recent advances in GPCR structural biology have produced hundreds of GPCR structures solved by X-ray diffraction and increasingly, cryo-electron microscopy (cryo-EM). Many of these structures are stabilized by site-specific cholesterol binding, but it is unclear whether these interactions are a product of recurring cholesterol-binding motifs and if observed patterns of cholesterol binding differ by experimental technique. Here, we comprehensively analyze the location and composition of cholesterol binding sites in the current set of 473 human GPCR structural chains. Our findings establish that cholesterol binds similarly in cryo-EM and X-ray structures and show that 92% of cholesterol molecules on GPCR surfaces reside in predictable locations that lack discernable cholesterol-binding motifs.

Advances in high-throughput molecular biology and electronic health records (EHR), coupled with increasing computer capabilities have resulted in an increased interest in the use of big data in health care. Big data require collection and analysis of data at an unprecedented scale and represents a paradigm shift in health care, offering (1) the capacity to generate new knowledge more quickly than traditional scientific approaches; (2) unbiased collection and analysis of data; and (3) a holistic understanding of biology and pathophysiology. Big data promises more personalized and precision medicine for patients with improved accuracy and earlier diagnosis, and therapy tailored to an individual\'s unique combination of genes, environmental risk, and precise disease phenotype. This promise comes from data collected from numerous sources, ranging from molecules to cells, to tissues, to individuals and populations-and the integration of these data into networks that improve understanding of heath and disease. Big data-driven science should play a role in propelling comparative medicine and \"one medicine\" (i.e., the shared physiology, pathophysiology, and disease risk factors across species) forward. Merging of data from EHR across institutions will give access to patient data on a scale previously unimaginable, allowing for precise phenotype definition and objective evaluation of risk factors and response to therapy. High-throughput molecular data will give insight into previously unexplored molecular pathophysiology and disease etiology. Investigation and integration of big data from a variety of sources will result in stronger parallels drawn at the molecular level between human and animal disease, allow for predictive modeling of infectious disease and identification of key areas of intervention, and facilitate step-changes in our understanding of disease that can make a substantial impact on animal and human health. However, the use of big data comes with significant challenges. Here we explore the scope of \"big data,\" including its opportunities, its limitations, and what is needed capitalize on big data in one medicine.

Currently, time-consuming serial in vitro experimentation involving immunocytochemistry or radiolabeled materials is required to identify which of the numerous Rab-GTPases (Rab) and Rab-GTPase activating proteins (RabGAP) are capable of functional interactions. These interactions are essential for numerous cellular functions, and in silico methods of reducing in vitro trial and error would accelerate the pace of research in cell biology. We have utilized a combination of three-dimensional protein modeling and protein bioinformatics to identify domains present in Rab proteins that are predictive of their functional interaction with a specific RabGAP. The RabF2 and RabSF1 domains appear to play functional roles in mediating the interaction between Rabs and RabGAPs. Moreover, the RabSF1 domain can be used to make in silico predictions of functional Rab/RabGAP pairs. This method is expected to be a broadly applicable tool for predicting protein-protein interactions where existing crystal structures for homologs of the proteins of interest are available.

A fundamental and unsolved problem in biophysical chemistry is the development of a computationally simple, physically intuitive, and generally applicable method for accurately predicting and physically explaining protein-protein binding affinities from protein-protein interaction (PPI) complex coordinates. Here, we propose that the simplification of a previously described six-term PPI scoring function to a four term function results in a simple expression of all physically and statistically meaningful terms that can be used to accurately predict and explain binding affinities for a well-defined subset of PPIs that are characterized by (1) crystallographic coordinates, (2) rigid-body association, (3) normal interface size, and hydrophobicity and hydrophilicity, and (4) high quality experimental binding affinity measurements. We further propose that the four-term scoring function could be regarded as a core expression for future development into a more general PPI scoring function. Our work has clear implications for PPI modeling and structure-based drug design.