RESEARCH-PGR: Predicting Gene-Specific Functional Contributions to Maize Reproduction: A Machine-Learning Approach

From a biological perspective, the success of flowering plants is intimately tied to the process of sexual reproduction, facilitated by flowers and their products, the male pollen and the female embryo sac embedded within the flower. In agriculture, primary products such as grain and fruit rely on the success of sexual reproduction. In particular, the ability of pollen to transmit sperm cells to the embryo sac is critical to the production of the next generation of seeds.

In maize, the focus of this project, environmental stressors such as high heat can lead to loss of pollen viability and subsequent crop disruption. This project will utilize a novel imaging and automated computer vision system (EarVision) coupled with a large set of easily assayed mutants to measure the contributions of hundreds of genes to maize pollen function.

Quantitative and computational approaches will analyze available genome-scale data to help predict gene function during maize reproduction and build an improved understanding of the pollen genetics of this crop plant. The project will inform approaches directed towards multiple agricultural goals such as improving crop resilience or controlling germ line activity.

In addition, the project will educate students at high school, undergraduate, and postdoctoral levels in plant genetics and quantitative scientific skills via outreach through summer camps and inclusive research projects.

This project targets the two cell types of the male gametophyte, vegetative and sperm cells, for a data-driven, systems-level investigation of genes underlying their biological functions using mutational interrogation as a test of gene function. Taking advantage of the haploid nature of pollen and a large set of fluorescently-marked insertional mutations known as Ds-GFP lines (acdsinsertions.org), the project will employ an innovative automated phenotyping approach to generate measures of gene-specific contributions to pollen fitness for several hundred genes.

Machine learning and statistical approaches will be used to analyze the resulting quantitative dataset, developing integrative models to relate omics-scale datatypes, such as genome, transcriptome, and proteome data, to the loss-of-function phenotypic outcome for specific genes. The optimal model will be used to predict pollen phenotype from genotype, and subsequently be validated via indirect and direct tests.

The project will also develop appropriate statistical methods to exploit the high-content images generated via the phenotyping system, to address hypotheses relating how altered pollen function can influence patterns in the progeny population on the maize ear inflorescence. The project brings together three Oregon State University faculty members with complementary expertise to support its transdisciplinary approach: in plant genetics and development, in statistics and systems analysis, and in machine learning and computational biology.

The project will train a postdoc in transdisciplinary application of computational and quantitative approaches to a biological question as well as undergraduate researchers in either genetic or quantitative approaches in biology in collaboration with the Oregon State STEM Leaders Program.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.