Dissertation Defense Announcement
To:  The George Mason University Community

Candidate: Jennifer Woodward-Greene

Program: PhD in Bioinformatics & Computational Biology



Date:   Tuesday April 19, 2016

Time:   2:00 PM
Place:  George Mason University
             Science & Tech (Prince William) Campus<http://www.gmu.edu/resources/welcome/Directions-to-GMU.html>
            Occoquan Bldg., Room 204





Title: "Collection and Recording Digital Phenotype and Body Weight Prediction Software"
Committee Chair: Dr. Iosif Vaisman
Dissertation Director:  Dr. Curt Van Tassell
Committee Members:  Dr. Tad Sonstegard, Dr. Jason Kinser

A copy of the dissertation is available in the Gateway Library.  All are invited to attend the defense.
ABSTRACT:
The United Nations Food and Agriculture Organization (FAO) affirmed accurate, consistent phenotypic data is integral to understand and conserve indigenous animal genetic resources (AnGR). Body weight is a critical phenotype for animal health, production, breeding, and marketing. Platform (walk-on) scales are the most accurate tool, however, they are expensive and inconvenient. Alternative, cheaper methods to predict body weight manually have been in common usage for decades. Current methods use cloth tape to measure chest (heart) girth around the rib cage, and body length. Caprine weigh tape (tape), measures chest girth and provides a corresponding body weight; while 'BM' predicts body weight in pounds as body weight in pounds  = ((chest girth inches  × chest girth inches × body length inches)/300). A novel collection protocol, calibration signs, and two digital phenotype enumeration software methods were developed as an alternative to manual body measurement weight prediction, with electronic phenotype recording. The ADAPTMap digital method is a semi-automated image segmentation and measuring algorithm. It improves the current state-of-the-art GrabCut image segmentation by 19.5%. The Points Method obtains digital measurements via user mouse clicks on the animal image. Digital body measures were highly correlated to manual measures (from 0.72 to 0.82 Pearson Correlation Coefficients (PCC)). ADAPTMap digital-only measures included total body area, and trunk area (body excluding head and legs). Digital body weight prediction models fit linearly (r2 = 0.90 and 0.83 for ADAPTMap and Points). Validation and field studies compared digital (ADAPTMap, Points) and existing (tape, BM) weight predictions to scale weights. The validation study included 77 goats, with calibrated, and certified platform scales. Physical body measures were collected by the same person for all 77 subjects, and digital color images were collected per the ADAPTMap protocol. The field study collected body measures and scale weights using several different sampling teams on 817 goats from five African countries. Field weights were measured on portable hanging scales; weight and body measures were recorded manually. Comparing scale weights to weight predictions showed PCCs of 0.86 and 0.91 (tape, BM) and 0.91 and 0.86 (ADAPTMap, Points). The digital and manual methods range of percent body weight difference between the scale and the predicted weights were similar for BM, Points, and ADAPTMap ([0% - 72%], [0% - 72%], [0% - 60%]), and larger for tape ([0% - 181%]). Despite tape and BM common usage on farms for individual animals, these ranges and PCCs, indicate both existing, and new digital methods are more suited for populations. The digital methods, as with many novel software and algorithms, are more likely to improve than tape and BM. The ADAPTMap and Points digital phenotype measures improve existing methods. They add digital only phenotypes, seamlessly provide digital records, and predict body weight similarly to current methods.

###