Rapid and automated lipid profiling by nuclear magnetic resonance spectroscopy using neural networks

Abstract

Nuclear magnetic resonance (NMR) spectroscopy is a powerful tool for quantitative metabolomics; however, quantification of metabolites from NMR data is often a slow and tedious process requiring user input and expertise. In this study, we propose a neural network approach for rapid, automated lipid identification and quantification from NMR data. Multilayered perceptron (MLP) networks were developed with NMR spectra as the input and lipid concentrations as output. Three large synthetic datasets were generated, each with 55,000 spectra from an original 30 scans of reference standards, by using linear combinations of standards and simulating experimental‐like modifications (line broadening, noise, peak shifts, baseline shifts) and common interference signals (water, tetramethylsilane, extraction solvent), and were used to train MLPs for robust prediction of lipid concentrations. The performances of MLPS were first validated on various synthetic datasets to assess the effect of incorporating different modifications on their accuracy. The MLPs were then evaluated on experimentally acquired data from complex lipid mixtures. The MLP‐derived lipid concentrations showed high correlations and slopes close to unity for most of the quantified lipid metabolites in experimental mixtures compared with ground‐truth concentrations. The most accurate, robust MLP was used to profile lipids in lipophilic hepatic extracts from a rat metabolomics study. The MLP lipid results analyzed by two‐way ANOVA for dietary and sex differences were similar to those obtained with a conventional NMR quantification method. In conclusion, this study demonstrates the potential and feasibility of a neural network approach for improving speed and automation in NMR lipid profiling and this approach can be easily tailored to other quantitative, targeted spectroscopic analyses in academia or industry.

Explainable AI to Facilitate Understanding of Neural Network-Based Metabolite Profiling Using NMR Spectroscopy

“…A spectrum of hepatic lipids extracted from a Nile Grass rat obtained in a prior study [ 15 ] was similarly assessed with the IG algorithm. A more detailed description of the sample preparation, data acquisition, and model training can be found in the previous manuscript, which validated this lipid-profiling MLP model [ 14 ] (MLP-3 is the specific model and Mix 2 is the specific lipid mixture used for this study). Attribution results for lipids are assessed primarily visually, with peaks deemed important by the IG method compared both to lipid standard signals ( Figure 2 ) and to resonances used by human spectroscopists for lipid quantification ( Supplementary Figure S1 ) [ 5 , 15 ].…”

“…(Elysian, MN, USA): tridocosahexaenoin [TriDHA], trilinolein, triolein, tripalmitin, cholesterol, methyl eicosapentaenoate [mEPA], palmitic acid, cholesteryl linoleate, and cholesteryl arachidonate, Tokyo Chemical Industry Co., (Tokyo, Japan):1,2-dipalmitoylsn-glycero-3-phosphocholine, 1,2-dioleoyl-sn-glycero-3-phosphocholine, 1,2-dipalmitoyl-snglycero-3-phosphoethanolamine, 1,2-dimyristoyl-sn-glycero-3-phosphoethanolamine, Avanti Polar Lipids, Inc. (Alabaster, AL, USA): 1-palmitoyl-2-hydroxy-sn-glycero-3-phosphocholine, and Matreya LLC (State College, PA, USA): sphingomyelin [bovine]). The standards were weighed individually to prepare 30 NMR samples (2 samples per standard) at various concentrations, and 1 H-NMR scans of these 30 samples obtained using a 400-MHz JEOL ECZ spectrometer (JEOL Ltd., Tokyo, Japan) were used to train an MLP for lipid quantification similar to MLP-Exp above (i.e., Adam optimizer, 200-node hidden layer, MSE loss function, and using a training dataset utilizing experimental variations and additional non-analyte signals [tetramethylsilane, water, and random singlets]) [14]. A spectrum for each lipid reference standard is shown in A mixture spectrum containing 13 lipid standards (all standards used in training except 1-palmitoyl-2-hydroxy-sn-glycero-3-phosphocholine and SM) was evaluated with the IG algorithm using a baseline spectrum of only the quantitative reference substance (dimethyl sulfone [DMSO2]) and NMR solvent (deuterated chloroform, deuterated methanol, and deuterium oxide-16:7:1-v/v/v).…”