Using Support Vector Machines to Model and Classify Phenotypic Differences in Diverse Maize Root Systems

This project was conducted as part of the Data-Driven Agriculture Research Experience & Extension for Undergraduates under the Department of Tropical Plant & Soil Science at the University of Hawaiʻi at Mānoa. The program is funded by the United States Department of Agriculture. In this project, shoot and root data was collected and analyzed for eight different maize inbred lines. This project has been presented at the University of Hawaiʻi at Mānoa’s Summer Undergraduate Research Experience (SURE) Symposium on Jul 30, 2021.

In the first half of this summer research internship, leaf and root data were collected. In the second half, we analyzed the data that we have collected. The data analysis is divided into four different sections: image analysis of the shoots, SVM analysis of the shoots, image analysis of the roots, and SVM analysis of the roots. This poster/oral presentation focuses on using Support Vector Machines to analyze root data. This presentation was conducted with Livvy Johnson of Iowa State, as a part of the bigger project in the REEU program. Other students who were involved in the bigger project are Bryceson Tugade, Lydia Rigge, Elizabeth Swantek, Alycia Tausaga, Jesse Mikasobe‐Kealiinohomoku, and Katie Strachan. The mentors for this project were Tai McClellan Maaz, Michael (Mikey) Kantar, Yuriy Mileyko, Michael Muszynski, and Nhu Nguyen.

Abstract: Maize is one of the most important global food crops accounting for nearly 30% of global calories. Not only is maize central to food security, it is also a model species for understanding plant growth and development. Using morphometric and image analysis of a diverse collection of maize inbred root systems, we applied support vector machines to classify different genotypes based on the combination of phenotypes. We will present results of our imaging and multivariate analysis to show how our models can provide classification insights not only in our eight tested genotypes, but also in unknown future genotypes. Results from this project can reveal how differences in root morphology among genotypes may influence nutrient acquisition patterns, resource use efficiency, carbon storage, and plant protection mechanisms. Ultimately, being able to classify unknown genotypes may help with the ability to place them in the optimal environment or agroecological system.