Developing an NIRS Prediction Model for Oil, Protein, Amino Acids and Fatty Acids in Amaranth and Buckwheat

Abstract

Amaranth and buckwheat are two pseudo-cereals preferred for their high nutritional value, are gluten free and carry religious importance as fasting food. Germplasm resources are the reservoir of diversity for different traits, including nutritional characteristics. These resources must be evaluated to utilize their potential in crop improvement programs. However, conventional methods are labor-, cost- and time-intensive and prone to handling errors when applied to large samples. NIRS-based machine learning to predict different nutritional traits is applied in different food crops for multiple traits. NIRS prediction models are developed in this study using the mPLS regression technique for oil, protein, fatty acids and essential amino acid estimation in amaranth and buckwheat. Good RSQ external (power of determination) values were obtained for the above traits ranging from 0.72 to 0.929. Ratio performance deviation (RPD) value for most of the traits ranged between 2 and 3, except for valine (1.88) and methionine (3.55), indicating good prediction capabilities in the developed model. These prediction models were utilized in screening the germplasm of amaranth and buckwheat; the results obtained were in good agreement and confirmed the applicability of developed models. It will enable the identification of a trait-specific germplasm as a potential gene source and aid in crop improvement programs.

Keywords

• machine learning
• RSQ
• RPD
• mPLS
• WINISI
• validation