Conceptual validation of self-organisation studied by spectroscopy in an endosperm gene model as a data-driven logistic strategy in chemometrics

Chemometric models including data pre-treatment and inspection software are able to evaluate the coarsely grained “top down” observational data from biological tissues through spectroscopic sensors in a dialogue with fine-grained analytical “bottom up” data in a sequential exploratory selection strategy. It facilitates interpretation and development of new theoretical concepts. This is demonstrated by a data set of 11 barley mutant endosperm genes and crosses focusing on the extremely high mutant gene discrimination capability by Near Infrared Reflection Spectroscopy (NIRS). It represents patterns of intact chemical bonds from self-organised endosperm tissues, which are validated to prior spectroscopic, chemical and genetic knowledge. Principal Component Analysis (PCA) of spectra could classify different endosperm mutant genes and changed gene backgrounds. A new carbohydrate pathway regulation from starch to β-glucan was identified. Specific mutant genes were defined visually by directly inspecting and validating spectral patterns from genetically defined barleys. Genetic concepts such as “the phenome” and “pleiotropy” were given new definitions, phenomenologically expressed as log1/R MSC (Multiple Scatter Corrected) NIRS fingerprints comparing mutants in an iso-genic background. Chemometric models are efficient in over-viewing genetic and environmental spectral differences but are not able to reproduce and predict in detail the finely tuned spectra from the “natural computer” of the self-organising tissue. Thus it is always necessary to return to spectral data for a final visual evaluation. In statistics and chemometrics mathematical formalism in data modelling should be integrated with the chemical and biological meaning of data-patterns. Ideas for problem solution may be combined from both sides.