Trait Prediction using ML, DL and other novel methods

With the recent success of using advanced machine learning and deep learning, across various domains and problems, several new open-source tools and technologies with various implementations are now available. One of the use cases that we are exploring within Plant Breeding at Bayer Crop Science team is to use genotypic data, environmental data and other relevant databases (such as management data) with deep architectures and/or other newer methods understand the plant response (yield, primarily) as function of the genotypic and environmental variables and build models that can capture the interactions between them. The goals of the candidates working in the project will be to comprehensively evaluate various machine learning methods to predict a single or multiple traits of choice. We will initially provide a data set containing the genomic data and the phenotypic data obtained through large scale field experiments across hundreds of locations. When combined with publicly available weather data at these locations, we have a rich data set with several layers of information. These methods will additionally be validated against the publicly available curated dataset from the Genome2Fields (G2F) initiative, to establish generalizability. Being able to predict the plant response has a significant impact on the future of plant breeding and agriculture in general. We can meet the diverse needs of growing populations by creating new varieties of seed faster and with fewer resources, like land and water, leading to a global food security and sustainability.