Machine Learning Algorithms for Functional Genomics: Soybean Genomics for Pathogen Resistance

The ever increasing availability of large scale genomic datasets urge the development of more informative data mining techniques. With extensive advancements in computation speeds and efficiencies, the capacity for large scale data analysis is transcending the hands of computer experts and becoming readily available to everyday bench and field scientists. Nevertheless, computer scientists and engineers have had amazing advancements in analysis tools and algorithms over the years, but these algorithms are frequently lost in translation, scarcely making it into the toolbox of everyday biologists. An example of this is a subset of artificial intelligence: Machine Learning, or conferring computers the ability to learn from data. Although machine learning has been studied for decades, it was until recently that the floodgates of genomic data were opened. This presentation aims to deliver a practical understanding of machine learning techniques that can be utilized in functional genomic approaches to biological problems. In particular, the use of Neural Networks for mapping of pathogen resistance genes in a recombinant inbred line (RIL) population of soybean is explored. By employing a vector of RILs, each composed of a vector of single nucleotide polymorphism (SNP) markers, and a vector of phenotypic values such as soybean cyst nematode (SCN) incidence, a neural network is shown to learn epistatic interactions essential for disease resistance, and predict whether an unseen RIL will be resistant or susceptible to SCN. Finally, the potential of machine learning to circumvent massive Plant Introduction phenotyping trials is explored.