Predicting the Genotype to Phenotype Relationship in Plants using Machine Learning and Deep Learning
Date
2023-03-28
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0002-5193-4415
Type
Thesis
Degree Level
Masters
Abstract
With advancements in science and technology, it has become easier and faster to perform complex computations in a matter of a few minutes. Combined with the evolution in the area of artificial intelligence, machine learning has taken the world by storm. The availability of programming libraries with built-in machine learning methods has made it commonplace to use machine learning in order to solve tasks from all fields of science. The economical and computational feasibility to generate dense genome-wide markers has also grown because of advancements in marker and genotyping technology. An amalgamation of these has opened up the potential to improve the process of agricultural development. Indeed, the tasks of genomic prediction and genomic selection can be performed swiftly and offer the potential to reduce the cycle time in plant breeding by making accurate phenotypic predictions for the crops to be grown.
In this study, we compare and evaluate the performance of computational models for the tasks of genomic prediction and genomic selection on four publicly available datasets of different species of agricultural crops. We examine and quantify the capability of various computational models to accurately predict the phenotypic values and determine the top-ranked samples to be used for the next breeding cycle.
We also look at two methodologies to determine the important genomic markers, and compare their performance. We found that convolutional neural network models based on different architectures were able to make the best predictions and were capable of solving the tasks of genomic prediction and genomic selection. We also found that the entropy-based methodology performed well for the task of determining the important genomic markers, and it aided the computational models in achieving higher prediction accuracy.
Finally, we created and presented a web application built to solve the tasks of genomic prediction, genomic selection, and marker assessment in one place. The web application is aimed at solving the discussed problems in an easy and intuitive manner by users.
Description
Keywords
artificial intelligence, machine learning, deep learning, prediction models, phenotype prediction, genomic prediction, genomic selection, marker selection, genotype phenotype relationship, curse of dimensionality, web application
Citation
Degree
Master of Science (M.Sc.)
Department
Computer Science
Program
Computer Science