What is NMR-DB?

NMR-DB is an online database of annotated NMR datasets, which can serve as a basis for development and validation of computational methods for protein NMR spectroscopy. Each dataset, stored in the repository, represents a single protein. In particular, it contains all NMR spectra recorded to solve the protein structure, along with their manual analysis (e.g. signal positions, chemical shifts and structure). NMR-DB could be utilized to establish data-driven approaches for end-to-end protein structure solving with NMR spectroscopy, as well as to evaluate or compare existing algorithms. This repository has been developed to address unsatisfactory situation, wherein analysis of protein NMR spectra is performed manually and automation of this process with advanced computational methods is hindered by lack of benchmarks.

Why NMR-DB is needed?

NMR spectroscopy is one of the leading techniques for protein structure elucidation. It has remarkable analytical capabilities, which allow solving a protein structure and studying its dynamics in its nearly-physiological conditions (without crystalization or freezeing the sample). A large obstacle in structural studies with NMR spectroscopy is elaborative data analysis process. Nowadays it takes months of manual work to derive structure from series of recorded NMR spectra. Elimination of this bottleneck is highly desired, but the problem remains unsolved for over 20 years. One of the important technical problems in this research field is lack of publicly available benchmark dataset. Establishing such benchmarks for mature computer science problems (i.e. object detection) were beneficial and lead to models performing respective tasks automatically with human-level performance. We believe that NMR-DB can largely simplify development of computational methods for protein NMR spectroscopy in the near future.