“…A more general approach is to model the spectra in an untargeted and direct way through optimization. , In particular, time-domain signal modeling directly tackles the overlap problem and is advantageous for nonuniformly sampled data and data with baseline problems. − The utility of time-domain modeling is demonstrated by CRAFT, which is the complete reduction to amplitude frequency table method. CRAFT, which is based upon a Bayesian framework, has been used to decompose 1D and 2D NMR spectra by time-domain modeling, and while CRAFT workflows can involve interactive partitioning of spectral regions, CRAFT performance is robust with respect to any user-defined parameters, and so it does have the potential for complete automation. − Similar ideas have been tested by researchers including Rubtsov et al, but extensive validation and consistent workflows are not available. − Singular value decomposition (SVD) and linear prediction-based methods provide alternative solutions, but the tuning parameter, namely, the reduced order, is hard to select for real-world data sets. − Alternatively, global spectral deconvolution (GSD) provides a convenient commercial solution that automatically fits frequency-domain data, but in crowded regions, it tends to overfit by using many non-ideal and nonphysical peaks. Therefore, an automated untargeted method to decompose NMR spectra is still needed.…”