Elements: Building Large-Language Model-Based Data Infrastructure for Enzyme Biotechnology

Computational enzyme engineering has the potential to accelerate the discovery of catalysts for sustainable manufacturing, medicine, biotechnology, and environmental applications. However, progress is limited by the fragmented state of enzyme data. These data are often scattered across publications, supplementary files, figures, and separate databases. They are difficult to compare, reuse, or analyze at scale, especially for researchers who lack specialized programming or database expertise. This project develops Mutexa.Data, an open and community accessible data infrastructure that transforms scattered enzyme knowledge into reliable, searchable, and usable resources. The project strengthens the data foundation for artificial intelligence assisted enzyme discovery and enables research relevant to sustainable biomanufacturing and biotechnology. The project also supports education and workforce development through web-based tools, tutorials, workshops, and integration into undergraduate and graduate courses, enabling students, instructors, and researchers to unlock the full potential of enzymology database through intuitive natural language interfaces. The project has two aims. The first aim is a large language model-based data extractor that inspects, validates, and enriches enzyme information from scientific literature and existing databases. This extractor uses document back checking, citation grounded verification, multi-agent error detection, sequence alignment, cross database normalization, and recognition of chemical information in figures to connect enzyme data with verified sequences, reactions, structures, and experimental metadata. It expands beyond standard kinetic constants to include broader properties such as activity, fitness, yield, selectivity, thermostability, solubility, inhibition, and assay conditions in a unified schema with quantified provenance and confidence. The second component is a large language model-based data navigator that allows users to query, clean, analyze, and interpret enzyme data through natural language. This navigator translates user questions into structured database queries, performs statistical and machine learning analyses, reports data quality information, and supports hypothesis driven reasoning grounded in literature evidence. Together, these components convert static enzyme databases into an interactive cyberinfrastructure for reliable data access, reproducible analysis, and artificial intelligence ready dataset generation. The platform is distributed through open-source software, public web interfaces, application programming interfaces, and community governed development, providing a sustainable foundation for enzyme engineering research and education. This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Directorate of Biological Sciences. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.