SAND: Automated Time-Domain Modeling of NMR Spectra Applied to Metabolite Quantification

Area of Science:

• Analytical Chemistry
• Biochemistry
• Computational Biology

Background:

• Untargeted nuclear magnetic resonance (NMR) metabolomics generates vast datasets from thousands of biological samples.
• Accurate quantification of spectral features is crucial for data exploitation but hindered by significant signal overlap.
• Existing workflows lack consistency and automation, posing challenges for large-scale metabolomics studies.

Purpose of the Study:

• To introduce Spectral Automated NMR Decomposition (SAND), a novel software for automatic and accurate quantification of NMR spectra.
• To address the challenge of signal overlap in untargeted metabolomics through an automated workflow.
• To enable detailed spectral feature quantification for enhanced data analysis in metabolomics.

Main Methods:

• Development of the Spectral Automated NMR Decomposition (SAND) software utilizing time-domain modeling.
• Implementation of hybrid optimization with Markov chain Monte Carlo methods and time/frequency domain subsampling.
• Validation using simulated, mixture, and real-world biological samples (urine) with varying complexity.

Main Results:

• SAND achieves high accuracy, with a correlation of approximately 0.9 with ground truth data.
• The software successfully quantifies spectra from highly overlapped simulated data, mixtures, and biological samples.
• Automated annotation using correlation networks demonstrates recovery of 74% of compound peaks on average.

Conclusions:

• SAND provides a robust, automated solution for NMR spectral quantification, overcoming signal overlap.
• The software enhances the utility of untargeted metabolomics by enabling detailed and consistent spectral analysis.
• SAND's time-domain subsampling approach offers potential for extension to higher dimensional and non-uniformly sampled data.