Applications and Evaluation of Phenomic Technologies in Maize Breeding
Abstract
High-throughput phenotyping technologies can generate large volumes of data at low costs and are useful in a wide range of plant sciences, including plant breeding. Many phenotyping technologies increase the rate of currently measured traits, while others can detect completely novel traits yet to be fully explored. In addition to replacing current manual measures, higher throughput technologies may be used to indirectly predict yield. Here, we explore this concept, using high-throughput phenotype information from Fourier Transformed near-infrared reflectance spectroscopy (NIRS) of harvested kernels to predict parental grain yield in maize (Zea mays L.). A dataset of 2,563 kernel samples from a diversity panel of 346 hybrid testcrosses were scanned using an FT-NIRS interferometer, measuring 3,076 wavenumbers (bands) in the range of 4,000- 10,000 cm-1 . Corresponding grain yield for each sample was used to train predictive models using three types of statistical learning: (a) partial least square regression (PLSR), (b) NIRS best linear unbiased predictors (NIRS BLUP) and (c) functional regression. These results found NIRS spectral data to be a useful tool in predicting maize grain yield of unknown samples and showed promising results for evaluating genetically independent breeding populations. High correlations between predicted and observed values demonstrated value for grain NIRS in ranking variety yields relative to one another, even where yield predictions were not accurate. More research in this area will provide better understanding for how NIRS and other phenomic technologies can be used to predict phenotypes in breeding programs.
The second project of this thesis explored the tradeoffs that exist between using high throughput phenotyping technologies with a higher error rate compared to a traditional phenotyping method with higher accuracy, with lower sampling rates. To accomplish this, populations were simulated to include variance from genetic, environmental, other effects, and error; different analysis scenarios were then created and assessed. It was determined that additional environmental sampling could compensate for greater measurement error; while for genetic mapping, increasing population size was most important to correctly predict genetic loci. This analysis provided an understanding of the threshold for high-throughput sample number increases needed to obtain a benefit over more accurate traditional measures.
Citation
Lane, Holly Marie (2019). Applications and Evaluation of Phenomic Technologies in Maize Breeding. Master's thesis, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /189088.