b'Machine Learning and Computer Vision is Bolstering BreedingA revolution in soybean breeding for root traits has begun, with the presentation of a successful new pipeline involving machine learning and computer vision.Treena HeinWHEN CROP BREEDERSlongtolerant root systems, using genomic ago learned of single nucleotideprediction-based approaches, says Falk, polymorphismsSNPs, differences in aintegrated field scientist for Corteva single building block/nucleotide such asAgriscience.cytosine in place of thymine, in a givenMy research was the first step to stretch of DNAthey have wanted to bemaking this happen but more work needs able to correlate these differences withto be done, specifically continuing this plant traits, taking breeding to an entirelywork into the field environmentwe new level.know that lab or greenhouse based Now, building on past achieve- research often doesnt align with what ments in the field, a team of scientistsgoes on in the field. Drought has hit the in the departments of Agronomy andwestern Canadian soybean grower hard Mechanical Engineering at Iowa Statein the last few years, and the best way University and at the U.S. Department offorward is drought-tolerant soybeans. But Agriculture (USDA) Agricultural Researchas with all of plant breeding, even with Service have published results on theirgenomic technology, its going to be a new soybean breeding framework thatlong haul.links genetic information with root traits using computer vision andmachineRoot System Architecturelearning (ML) tools. Because RSA affects water/nutrient Project co-lead Asheesh K. Singh, pro- acquisition, plant-microbe interaction, fessor in the Department of Agronomy atnutrient storage and structural anchor-Iowa State University, notes that some ofage of a plant, it can greatly impact yield. these methods have already been usedSoybean taproots that elongate at a with tremendous success to measurefaster rate tend to burrow more deeply crop traits or predict yield.PhD student Clayton Carley is working oninto the soil profile, possess increased More mainstream deployment hasusing deep learning coupled with imageroot densities at depth, and can therefore happened in the past five to six years, heprocessing tools to measure above-groundbetter access water in dry soils. However, explains. However, in terms of the studytraits that could be correlated with soybeanfor optimum plant performance, these of roots for plant breeding applications,root traits.deep roots need to be complemented our project is among the first to establishwith shallow lateral roots which enable a user-friendly pipeline built on ML toolssystem architecture (RSA) to the genomethe plant to efficiently forage for soil-and then use the pipeline to study traits inof the plants. Those plants with RSA thatimmobile nutrients. At this point in time, a way, and at a scale, that was previouslyenables greater drought tolerance andmore research is needed to understand difficult. nutrient acquisition are of obviously of par- the optimum RSA in a broad range of Manitobas Kevin Falk, who wasticular interest in breeding for better plantgrowing conditions. Singhs PhD student, and Zaki Jubery,performance and higher yield. In any case, although RSA can greatly scientist in the group of co-lead BaskarRoot phenotype combined with SNP- impact yield, crop breeding programs Ganapathysubramanian, and their col- based genotype allowed us to correlatehave rarely used it as a direct selection leagues imaged thousands of soybeangenetics with morphology. Its a firstcriterion. Instead, RSA has been indirectly root systems and used ML to measurestep that could allow breeders to selectselected for in plants that perform well in root traits and correlate differences in rootfor root morphologies, such as drought- general and/or perform well under dry 108/ SEEDWORLD.COMDECEMBER 2020'