Advancing Complex Phenotype Analyses through Machine Vision and Computation

University Of Wisconsin-Madison, Madison WI

Investigators

Abstract

PI: Edgar P. Spalding (University of Wisconsin) Co-PIs: Tessa L. Durham Brooks (Doane College), Nicola J. Ferrier (University of Wisconsin), Nathan D. Miller (University of Wisconsin) and A. Mark Settles (University of Florida) Collaborators: Paul Armstrong (USDA-ARS) and Gokhan Hacisalihoglu (Florida Agricultural and Mechanical University) One way to learn what a particular gene does for an organism is to determine what happens when the gene is eliminated, or mutated. In genetic terms, such a biological effect of a mutation is a phenotype. The central thesis of this project is that techniques for finding and understanding phenotypes are far less advanced than those for cataloging and manipulating the genes. By integrating engineering and computer sciences methodologies, this project will have special impact on the problem of detecting and quantifying plant phenotypes by increasing the types, precision, degree of automation, and throughput of measurements. This will increase the amount and quality of phenotype information that can be extracted from the burgeoning collections of mutants and systematically structured populations of naturally-occurring genetic variants. Specifically, the chemical composition of seeds, seed morphology, and the growth and behavior of the root that emerges after germination will be analyzed with spectroscopy and/or digital image analysis. The results will be subjected to quantitative modeling to search for relationships that have predictive power. One result to be expected, based on preliminary success, is that one phenotype can predict another. For example, the amount of oil in a corn seed inferred from reflected infrared spectra can predict something about the behavior of the root that develops after germination, as studied by time-lapse image analysis. Results like these will be important because 1) they give information about how individual genes in an organism interact to produce the myriad functions and behaviors, and 2) they may provide alternative means for plant breeders to develop varieties that have desired traits. It may be easier and faster to select for a particular root growth pattern when developing a new corn variety than directly measuring oil content, for example. Successful large scale phenotyping requires automated data acquisition and computation. Therefore, this project emphasizes both hardware and software development. Because people, not machines, drive science forward, the project includes plans for geographically separated experts to function as a cyber-enabled virtual organization. A distinguishing feature of this project is the necessary integration of engineering, computer sciences, biology, and different types of institutions, including two major research universities, an institution with a longstanding role in serving underrepresented groups, and a primarily undergraduate institution. Undergraduate students who participate in the project contribute fully to data analysis, acquisition and design. Only if this integration is successful will the scientific goals be achieved and the cyber-enabled ways of working in plant functional genomics tested. The results of the project including methods, equipment and software will be accessible at the project website (http://phytomorph.wisc.edu) and through the iPlant Collaborative.