Deriving Three One Dimensional NMR Spectra from a Single Experiment Through Machine Learning

NMR-based metabolomics is a robust analytical tool used to investigate the chemical phenotype of biofluids. The relationships between its components and human health has the potential to alter clinical practices, including treatment efficacy and disease prognosis. As an intrinsically quantitative, highly reproducible and non-destructive technique, often requiring little sample preparation, NMR is ideal for metabolomics. Utilizing targeted pulse sequences, it is possible to enhance or suppress certain molecular signals depending on molecular weight, leading to the selective measurements of small metabolites and macromolecules, both present in biofluids. Currently, the collection and analysis processes of NMR spectra together are time-intensive, which is inefficient for high-throughput studies. Here the authors show these processes can be simplified through machine learning predictions of three one-dimensional NMR spectra from one acquired one-dimensional NOESY spectrum per sample with PLS regression models. Using several statistical measurements, the authors found a high success rate for the predictions of both CPMG and diffusion-edited spectra with moderate success for pure-shift pJRES predictions. In order to further test each prediction, they verified each model on an individual test center’s samples, further underscoring the models’ predictions through correlation measurements. They used the predictions for further investigations into real-case scenarios. By demonstrating the predicted spectra’s ability to calculate concentrations of twenty common metabolites and then building random forest classification models for survival prognosis in acute myocardial infarction patients, the authors displayed the accuracy of this streamlined process. These results present significant advancements towards optimizing not only NMR spectral acquisition, processing, and multivariate analysis, but demonstrate the potential to implement machine learning-predicted NMR spectra in clinical practice.

References

Vignoli A, et al. (2025) Deriving three one dimensional NMR spectra from a single experiment through machine learning. Nature Communications 16,10159.

Xiao X, et al. (2024) Using neural networks to obtain NMR spectra of both small and macromolecules from blood samples in a single experiment. Communications Chemistry 7, 167.