Phenotype Classification of Intact Cells by NMR Spectroscopy through Machine Learning Approaches.

Area of Science:

• Biophysics
• Metabolomics
• Machine Learning

Background:

• Nuclear Magnetic Resonance (NMR) spectroscopy analyzes biological samples via metabolic fingerprinting.
• Current NMR methods struggle with live cells and in vivo imaging due to signal overlap from sample inhomogeneity.
• High-resolution NMR is effective for biofluids and cell extracts but limited for complex live systems.

Purpose of the Study:

• To investigate the use of machine learning for classifying live cell types using NMR spectroscopy.
• To overcome signal overlap limitations in NMR spectra of intact cells.
• To assess the potential for in vivo applications in diagnostics.

Main Methods:

• Applied machine learning algorithms to poorly resolved 1D NMR spectra of live, intact cells.
• Recorded NMR spectra at high magnetic fields.
• Trained classifiers on specific cell types (neural progenitor cells, neurons, astrocytes) and mixed cultures.

Main Results:

• Successfully classified distinct physiopathologically relevant cell types in vitro using machine learning on NMR data.
• Demonstrated classification of mixed cell type samples.
• Showed that classifiers trained at high fields can discriminate cells analyzed at lower fields, relevant to MRI.

Conclusions:

• Machine learning enables classification of live cell types from challenging NMR spectra.
• This approach holds promise for future Magnetic Resonance Spectroscopic Imaging (MRSI) data analysis.
• Potential for noninvasive diagnostics of central nervous system lesions, reducing biopsy needs.