The analysis of nuclear magnetic resonance (NMR) spectra for the comprehensive and unambiguous identification and characterization of peaks is a difficult, but critically important step in all NMR analyses of complex biological molecular systems. Here, we introduce DEEP Picker, a deep neural network (DNN)-based approach for peak picking and spectral deconvolution which semi-automates the analysis of two-dimensional NMR spectra. DEEP Picker includes 8 hidden convolutional layers and was trained on a large number of synthetic spectra of known composition with variable degrees of crowdedness. We show that our method is able to correctly identify overlapping peaks, including ones that are challenging for expert spectroscopists and existing computational methods alike. We demonstrate the utility of DEEP Picker on NMR spectra of folded and intrinsically disordered proteins as well as a complex metabolomics mixture, and show how it provides access to valuable NMR information. DEEP Picker should facilitate the semi-automation and standardization of protocols for better consistency and sharing of results within the scientific community.

A customized metabolomics NMR database, termed ¹H(¹³C)-TOCCATA, is introduced, which contains complete ¹H and ¹³C chemical shift information on individual spin systems and isomeric states of common metabolites. Since this information directly corresponds to cross sections of 2D ¹H–¹H TOCSY and 2D ¹³C–¹H HSQC-TOCSY spectra, it allows the straightforward and unambiguous identification of metabolites of complex metabolic mixtures at ¹³C natural abundance from these types of experiments. The ¹H(¹³C)-TOCCATA database, which is complementary to the previously introduced TOCCATA database for the analysis of uniformly ¹³C-labeled compounds, currently contains 455 metabolites, and it can be used through a publicly accessible web portal. We demonstrate its performance by applying it to 2D ¹H–¹H TOCSY and 2D ¹³C–¹H HSQC-TOCSY spectra of a cell lysate from E. coli, which yields a substantial improvement over other databases, as well as 1D NMR-based approaches, in the number of compounds that can be correctly identified with high confidence.

A new metabolomics database and query algorithm for the analysis of ¹³C–¹H HSQC spectra is introduced, which unifies NMR spectroscopic information on 555 metabolites from both the Biological Magnetic Resonance Data Bank (BMRB) and Human Metabolome Database (HMDB). The new database, termed Complex Mixture Analysis by NMR (COLMAR) ¹³C–¹H HSQC database, can be queried via an interactive, easy to use web interface at http://spin.ccic.ohio-state.edu/index.php/hsqc/index. Our new HSQC database separately treats slowly exchanging isomers that belong to the same metabolite, which permits improved query in cases where lowly populated isomers are below the HSQC detection limit. The performance of our new database and query web server compares favorably with the one of existing web servers, especially for spectra of samples of high complexity, including metabolite mixtures from the model organisms Drosophila melanogaster and Escherichia coli. For such samples, our web server has on average a 37% higher accuracy (true positive rate) and a 82% lower false positive rate, which makes it a useful tool for the rapid and accurate identification of metabolites from ¹³C–¹H HSQC spectra at natural abundance. This information can be combined and validated with NMR data from 2D TOCSY-type spectra that provide connectivity information not present in HSQC spectra.