Abstract

High throughput phenotyping is shifting plant genetics towards high(er) dimensional trait data sets. Once a set of images, point clouds, or hyperspectral reflectance measurements have been collected, the marginal cost of quantifying additional plant traits from the same sensor data is low. In an example of the utility of multidimensional trait, 3D reconstructions from 2D images and organ-instance segmentation can be used to map genes controlling leaf angle variation on a leaf-by-leaf level in sorghum, incorporating data
collected from the same plants at multiple time points. Community association populations which have been widely adopted by multiple research groups also produce high dimensional trait data, and these data can be used to identify pleiotropic effects of both
known mutants and previously uncharacterized loci. Curated datasets of more than 200 hundred traits were assembled in both maize and sorghum, and analyzed using both GPWAS and multivariate adaptive shrinkage. In maize, GPWAS predicts pleiotropic effects for loci which are consistent with loss of function phenotypes. In sorghum, MASHR identifies previously unknown effects of a classical dwarfing gene on root architecture. Given the value of high dimensional trait data from community association panels there is substantial value in structuring the collection of high throughput sensor data from association populations in ways that enable recycling and reuse.