NmrPipe1 is a widely used software package for the processing of NMR data that now includes NUS reconstruction capability. SMILE2 is an NUS reconstruction tool designed to be inserted into an nmrPipe processing pipeline. Since both packages work with data in nmrPipe format it is possible to use the same data to test both packages and compare the results.
Sampling schedules generated by the NUSscore3 software were used to extract unprocessed raw data rows from a fully sampled spectrum of strychnine in chloroform. Ten different sampling schedules at each of four different sampling levels (50, 25, 12.5 and 6.25%) generated forty different unprocessed data sets These datasets were then processed with identical parameters using both nmrPipe and SMILE. The processed spectra were evaluated for maximum and minimum intensity, and the noise estimated.
The graph below shows the mean noise of the spectra processed by nmrPipe (red) and SMILE (blue). Each point is the mean value from the ten spectra at each sampling level. The error bars indicate +/- one standard deviation. Spectra from both packages show the expected square root dependence4. Since noise is proportional to the square root of the acquisition time, removing some of the data (by sampling only 50,25,12.5 or 6.25%) results in less noise as the sampling is reduced.
The next graph compares the maximum intensity in the processed spectra. In this case the two packages produce quite different results. The nmrPipe spectra (red) maintain the same intensity over different amounts of sampling, but as the sampling is reduced the variation increases, as shown by the increase in the size of the error bars. In the SMILE spectra (blue) the maximum intensity shows a linear dependence on the amount of sampling. Unlike the nmrPipe spectra, the SMILE spectra do not show increased uncertainty at lower levels of sampling.
The difference in behaviour of the two packages is probably purposeful. It's likely the coders of the packages made different choices. In the case of the SMILE data the maximum intensity shows a linear dependence on the amount of sampling. Since signal intensity is proportional to acquisition time4, reduced sampling is equivalent to removing some of the acquisition time and thus reducing the signal. In the case of the nmrPipe data the intensity remains fairly constant and therefore the acquisition time must also be constant. With reduced sampling, however, parts of the acquisition time are not collected but are replaced with predicted data. The intensity in the nmrPipe data must be due to the combined experimental and predicted data, and this is why the uncertainty increases as the amount of reconstructed data increases. The decrease in the intensity of the SMILE data as the sampling is reduced indicates that the reconstructed data is not contributing to the signal intensity.
The final graph shows the maximum intensity divided by the noise. This is equivalent to the signal-to-noise ratio. The two packages show opposing behaviour. The nmrPipe spectra (red) show a counter intuitive increase in intensity-to-noise as sampling is reduced, while the SMILE spectra (blue) give a more expected decrease in intensity-to-noise with reduced sampling. This can be attributed to the differences seen in the second graph. In the case of the nmrPipe spectra, reduced sampling reduces the noise but keeps the signal intensity constant, resulting in an increase in intensity-to-noise as the sampling is reduced. For the SMILE spectra, reduced sampling reduces the noise with a square root dependence and reduces the intensity with a linear dependence. Dividing the two gives an intensity-to-noise ratio that decreases with a square root dependency on the amount of sampling.
This study shows that data processed in the same way, by the same algorithm, bur implemented by two different groups, can produce different results. This emphasises that it is difficult to compare the results of different software or to make general statements about the efficacy of algorithms. Furthermore, one needs to carefully examine and understand the data being analysed before drawing conclusions.
References
1. Delaglio F, Grzesiek S, Vuister GW, Zhu G, Pfeifer J, Bax A
NMRPipe: a multidimensional spectral processing system based on UNIX pipes.
J Biomol NMR. 1995 Nov;6(3):277-93.
2. Ying J, Delaglio F, Torchia DA, Bax A
Sparse multidimensional iterative lineshape-enhanced (SMILE) reconstruction of both non-uniformly sampled and conventional NMR data.
J Biomol NMR. 2017 Jun;68(2):101-118
3. Aoto PC, Fenwick RB, Kroon GJ, Wright PE
Accurate scoring of non-uniform sampling schemes for quantitative NMR.
J Magn Reson. 2014 Sep;246:31-5
4. Rovnyak D
The past, present and future of 1.26 T2
Concepts Magn Reson Part A Bridg Educ Res. 2019 Apr;47A(2). pii: e21473
No comments:
Post a Comment