Semester of Graduation
Fall 2023
Degree
Master of Science (MS)
Department
Biological Sciences
Document Type
Thesis
Abstract
Object detection is a recent development that allows for the detection of multiple objects in photographs. Numerous machine learning algorithms have been developed for object detection. Here we sought to determine which algorithm performed best when relating to an organism's health and immunological response. The fall armyworm (Spodoptera frugiperda) is a major agricultural pest causing millions of dollars in damage to farmlands across the globe. With their recent spread into Africa and Asia, the need for understanding how best to control population outbreaks via biocontrol has increased. While viruses are a major biocontrol agent for controlling insect pests, we still lack knowledge about how fall armyworm’s immune systems react when exposed to specialist pathogens and how that reaction varies with global warming. Current methods of measuring immune response use manual counting of hemocytes requiring large time investments. Here we use machine learning and compare several object detection algorithms on their ability to automatically count hemocytes. To perform counts, we trained four models on 398 photos and validated training on 114 photos. To gauge immunological responses under a changing climate, we reared fall armyworms generationally at two temperatures, 26° C and 31° C, infecting each generation with virus and allowing the survivors to reproduce. Hemocytes were extracted by cutting the fourth proleg, and hemolymph was mixed with anticoagulant for use in photographing on a hemocytometer. Of the algorithms compared, YOLOv8 was the most accurate and the quickest to train. Counts tended to be accurate and showed that the populations reared under colder temperatures had a higher immune response than those reared under warmer conditions. However, infected populations that coevolved with the virus showed no significant difference in immune response when comparing between temperatures. Fall armyworms in colder temperatures are better adapted to respond to virus due to the lack of heat response required. However, with exposure over multiple generations, temperature seems to have little effect on the fall armyworm’s immune response. This new development will drastically speed up the process of measuring insect immune response and open the door for further research on biocontrol methods in insects.
Date
11-2-2023
Recommended Citation
Haulk, Nathaniel T., "Using Machine Learning to Measure Changes in Immunity due to Climate Change in the Fall Armyworm, Spodoptera frugiperda" (2023). LSU Master's Theses. 5874.
https://repository.lsu.edu/gradschool_theses/5874
Committee Chair
Elderd, Bret D.