With the escalating persuasion of economic and nutritional importance of rice grain protein and nutritional
components of rice bran (RB), NIRS can be an effective tool for high throughput screening in rice
breeding programme. Optimization of NIRS is prerequisite for accurate prediction of grain quality parameters.
In the present study, 173 brown rice (BR) and 86 RB samples with a wide range of values were used
to compare the calibration models generated by different chemometrics for grain protein (GPC) and amylose
content (AC) of BR and proximate compositions (protein, crude oil, moisture, ash and fiber content)
of RB. Various modified partial least square (mPLSs) models corresponding with the best mathematical
treatments were identified for all components. Another set of 29 genotypes derived from the breeding
programme were employed for the external validation of these calibration models. High accuracy of all
these calibration and prediction models was ensured through pair t-test and correlation regression analysis
between reference and predicted values.
2015 Elsevier Ltd. All rights reserved.