Miniature (key-fob) NIR spectrometer results

Really interesting, thanks for sharing it Ian.
JM

It is interesting to see how people are trying to use miniature spectrometers to do something useful. These researchers have made some rather large assumptions in thinking that a single sample of 50 seeds represents a cultivar. Anyone with grain experience knows that tremendous variation can exist between samples of the same cultivar produced in different environments, locations or from different production seasons. Quantitation would have been a more interesting project as this would have implications for breeding line improvement and for utilization of these grains. "Fingerprinting" using NIR with such a small wavelength range is probably not the best use of this technology.

Thanks for this Ian!
Fred McClure would have loved it. I am impressed. Looks like a very interesting bit of kit.
And thanks for maintaining the Forum
Tony
 

Thank you Ian  for pointing the paper in my direction. Sadly, I must state up front that I am disappointed that this has been published because the conclusions cannot be justified from the data presented. While you were prompted to act because an SCIO instrument  was used for this study I am   more concerned with the inadequate application of chemometrics.
 
 I have seen several papers in recent years which attempt to report that NIR spectroscopy can be used to identify cultivars of grains crops, to determine if a grain is genetically modified, or if samples come from a particular region or country. At least one of the papers submitted  to JNIRS  is cited in this paper (it was not accepted for publication in JNIRS).  The main reason such papers were not accepted for publication in JNIRS is that the authors did not (attempt to) identify and explain the basis of the models they wished to report. Had they identified key wavelengths and related these to chemical bonds then there may have been justification for the conclusions (claims) they were making. On the other hand they may have discovered that the samples (cultivars,  GMO lines, or samples from different regions) only differed in moisture, protein, particle size, colour, or, as the authors state in their introduction “ The main sources of spectral variance include grain size… grain shape, and curvature (12) as well as grain position and orientation.”
 
The article in question published in PLoSone  in March 2018 should not have been published for the reason just given.  It should also have been rejected because

1. The  samples were not described in any detail. Did all 50 grains  come from a single spike?
2.  The presentation of samples to the instrument is poorly described and the statement “All grain samples were carefully scanned …Fig 1” is meaningless  How many scans made a spectrum? Was there background in the region of interest?
3. The chemometric techniques applied to the spectra, while powerful, produced models which were possibly overfitting for the 50 samples per cultivar.   Again I want to know more about the origin of the samples.
4.  To be realistic and in view of the topics covered in the Discussion section the study should have compared the ability of the SCIO instrument against a research grade instrument.   While the low cost SCIO  instrument may give useful results a more expensive instrument may give a better investment to a costly breeding program.

GB