Metodologia para distinção entre genótipos convencionais e transgênicos de algodão utilizando espectroscopia NIR e imagens hiperespectrais

Abstract

Genetically modified organisms (GMOs) are any organism in which the genetic material has not been altered in a natural way by mating or / and natural recombination. Genetic modifications are carried out to ensure, for example, that plants are tolerant to herbicides or resistant to fungi and insects, allowing for more robust varieties, of higher quality and yield. As GMOs have established themselves as the fastest technology adopted in the recent history of modern agriculture, some people, companies, agencies and governments have remained skeptical about the impacts that these organisms can cause on humans and the environment. In view of this, some laws have been developed in relation to the use, identification, quantification and differentiation of conventional cultivars from genetically modified ones. Currently, the conventional methods used for this purpose are laborious, expensive, time consuming and do not preserve the samples. Given the above, the objective of this work is to develop an alternative methodology to conventional ones for the quick and precise distinction between conventional cultivars and cotton GMOs using NIR spectroscopy (NIRS) and hyperspectral images (HSI). The adopted methodology involves the acquisition of the samples: 2390 seeds were analyzed in the NIR spectroscopy and 70 seeds analyzed by spectral images. For the construction of the PLS-DA model based on NIRS, the SP-Xy algorithm was used to select samples from the training and test sets. Then the spectra were pre-processed to remove spectral characteristics unrelated to the sample signal. The PLS-DA-NIRS model was then validated and applied to the samples in the test set. For the construction of the PLS-DA-HSINIR model, the images of 20 conventional and 20 transgenic seeds, obtained separately, were used in the training set, and for the test set, the image of 30 seeds randomly placed on the support was used. image acquisition. The image treatment was performed and the PLS-DA-HSINIR model built was designed on the training samples. The techniques used are non-destructive, do not use chemical reagents, fast and represent alternative methodologies that can help the classification / identification of transgenic cotton seeds. The PLS-DA-NIRS methodology presented an excellent predictive capacity, obtaining a correct classification rate of 97.95% in the prediction stage. The PLS-DA-HSINIR models uniquely classified 70% of the analyzed seeds. 7 of the samples had an inconclusive classification (23.3%). 2 conventional samples were classified incorrectly in the transgenic class representing a false positive. The constructed model shows the quality of correctly classifying 100% of the transgenic samples in their respective class, without the presence of false negatives. It can be concluded that the methodologies developed were able to distinguish transgenic samples from conventional ones, being therefore effective techniques for this purpose.