Proton nuclear magnetic resonance (NMR) helps identify molecules from hydrogen signals. Reported in August, SE2PSNet simplifies crowded spectra while preserving the peak areas used to quantify relative proton populations.[1]
The quantity is in the area
Quantitative NMR has an acquisition constraint that a neat-looking plot can conceal. Different proton signals can recover at different rates between measurements. If the next pulse arrives too soon, those signals need not recover equally, and their areas can cease to represent the underlying amounts faithfully. UCSB’s NMR facility explains the resulting trade-off between measurement time, signal strength and integral accuracy.[2]
Its guidance relates that choice to T1, the relaxation time describing recovery. A setting that works well for one signal may be inadequate for another in the same sample. Allowing more recovery time can improve quantitative accuracy, although it takes longer. That is a reason to examine the acquisition settings alongside any reconstructed spectrum.[2]
A test with three equal contributions
Trained on 41,000 simulated datasets, SE2PSNet uses several spin-echo spectra and a map of peak locations. In estradiol, three overlapping signals expected to contribute equally yielded relative integrals of 0.95:1:1.05. The established PSYCHE method gave 0.88:1:1.12. These are results for that region, not universal error rates.[1]
Our calculation shows deviations from one of 5% and 12%, respectively. Both ratios still sum to three: a correct total can conceal errors in individual contributions.[1]
In a hypothetical quality check, two reconstructed components could have areas of 1.5 and 0.5 when both should be one. Their combined area would still be two, despite a large error in the balance between them. Resolving the components is useful only when their individual quantities remain meaningful.
Higher digital resolution corrected a misidentified solvent signal. The result links better separation to better measurement: a useful reconstruction must preserve both the components and their relative amounts.[1]
Sources & further reading
- High-quality pure shift NMR spectra by deep learning using multi-spectral input and joint loss functions Nature Communications · 3 Aug 2026 · Accessed 20 Oct 2026
- Quantitative NMR: optimizing signal-to-noise and integral accuracy UCSB NMR Facility · Accessed 20 Oct 2026