Research Project
10267787
PROJECT SUMMARY/ABSTRACT This project continues the development of the UniProt Knowledgebase, which aims to provide the scientific community with a comprehensive, high-quality, and freely accessible resource of protein sequences and functional information. Proteins are an essential bridge between human genetics, the environment and phenotype. While human genetics has increasing power to find correlations between genotype and phenotype, knowledge of how proteins function, provided by UniProt, is essential for the mechanistic understanding critical to develop health outcomes through improved and personalized diagnostics, prognostics, and treatments. Biomedical research is being revolutionized by methods from the field of Artificial Intelligence, particularly Machine Learning (ML) approaches such as Deep Learning (DL). These approaches now outstrip the ability of humans in many fields and are state-of-the-art when sufficient data is available. UniProt provides gold standard training data for hundreds of ML applications in biomedical research. The work in this proposal will enhance the readiness of UniProt for use in ML and will integrate ML methods to enhance our efficiency. UniProt curators extract and synthesize experimental knowledge of proteins from papers in human and machine- readable forms using a range of standard ontologies. This proposal will further structure protein knowledge in UniProt, developing complete, machine-readable catalogs of the functional impact of human variation and of human protein networks and complexes, essential to understanding human disease. Efficiency of curation will be improved using DL models, developed in collaboration with text mining experts, to automate the identification of relevant papers and accelerate extraction of knowledge. This extracted knowledge will be validated by our expert curators and also the wider research community who will be actively engaged to further scale curation. ML approaches will also be used to infer annotations for proteins with no experimental characterization, using community challenges to develop faster, more accurate, scalable approaches to annotate the deluge of uncharacterized proteins. UniProt is an exemplar FAIR resource and has served the scientific community with metronomic data releases despite an exponential growth in data volumes. Streamlined production processes will scale efficiently and sustainably with both the growing data volume and complexity. We will explore novel technologies to ensure the continued timely release of data to the community according to the FAIR principles. UniProt is an international hub of protein data that serves hundreds of thousands of users annually. We will continue using user-centric approaches to develop the UniProt website in response to user needs and new data types. We will engage with our stakeholders and collaborators by introducing an annual strategic partnership meeting. We will engage our communities through webinars, social media, hackathons and attendance at scientific meetings to broaden the efficient and impactful use of our data.
