LOCAL Regression Algorithm Improves near Infrared Spectroscopy Predictions When the Target Constituent Evolves in Breeding Populations

Abstract

The CGIAR Harvest Plus Challenge Program began in the mid-2000s to support the genetic improvement of nutritional quality in various crops, including the carotenoids content of cassava roots. Successful conventional breeding requires a large number of segregating progenies. However, only a few samples can be quantified by high performance liquid chromatography each day for total carotenoids (TCC) and β-carotene (TBC) contents, limiting the gains from breeding. This study describes the usefulness of near infrared (NIR) spectroscopy and the efficiency of a large database coupled to a LOCAL regression algorithm to reach accurate TCC/TBC predictions on fresh cassava roots. The cassava database (6026 samples) was built over six years. TCC values ranged from 0.11 μg g−1 to 29.0 μg g−1, whereas TBC ranged from negligible values up to 20.1 μg g−1. All values were measured and expressed on a fresh weight basis. Between 2009 and 2014 increases in TCC and TBC were 86% and 122%, respectively. A comparison of calibrations using partial least squares (PLS) regression and LOCAL regression was done. The standard error of prediction were 1.82 μg g−1 for TCC and 1.28 μg g−1 for TBC using PLS model and 1.38 μg g−1 and 1.02 μg g−1, respectively, using LOCAL regression. The specificity of the data, with increasing content of the constituent of interest year after year, clearly showed the limitation of the classical partial least squares regression approach. The LOCAL regression algorithm takes advantage of large databases; this study highlighted the efficiency of this concept. NIR spectroscopy coupled to LOCAL regression led to efficient models for breeding programmes aiming at increasing carotenoids content in fresh cassava roots. NIR spectroscopy can also be used to predict other important constituents such as dry matter content and cyanogenic glucosides.