Machine learning reveals genes for nitrogen use efficiency in corn

Machine learning can identify important genes that help crops grow with less fertilizer, according to a new study published in Nature Communications.

“Now that we can more accurately predict which corn hybrids are best for the use of nitrogen fertilizers in the field, we can quickly improve on this characteristic. Increasing the efficiency of nitrogen use in corn and other crops offers three key benefits by reducing farmer costs, reducing environmental pollution, and mitigating gas emissions. greenhouse effect from agriculture, ”said the study’s author. Stephen moose, Alexander professor of Crop Sciences to University of Illinois at Urbana-Champaign.

Using genomic data to predict agricultural outcomes is both a promise and a challenge for biologists. Researchers are working to determine how to use vast amounts of genomic data to predict how organisms respond to changes in nutrition, toxins, and exposure to pathogens, which in turn would help improve crops. But the implications go beyond cultures, providing information on disease prognosis, epidemiology and public health.

However, accurately predicting complex agricultural and medical outcomes from genome-wide information remains a significant challenge.

As a proof of concept, the researchers demonstrated that machine learning models could predict genes important for nitrogen use efficiency in corn. A key first step was to find genes that respond to nitrogen in the leaves of field-grown corn plants and Arabidopsis, a small flowering plant widely used as a model organism in plant biology.

Nitrogen is an essential nutrient for plants and the main component of fertilizers; crops that use nitrogen more efficiently grow better and require less fertilizer, which has economic and environmental benefits.

“We show that focusing on genes whose expression patterns are evolutionarily conserved across species enhances our ability to learn and predict ‘genes of importance’ for the growth performance of staple crops, as well. that disease results in animals, “explained Gloria coruzzi, Carroll & Milton Petrie Professor in the Department of Biology and the Center of Genomics and Systems Biology at NYU and lead author of the article.

The researchers conducted experiments that tested whether eight “master switch” genes predicted from the machine learning model actually contribute to nitrogen use efficiency. They showed that altered expression of these switch genes in Arabidopsis or corn could increase plant growth in low nitrogen soils, which they tested both in the lab at NYU and in corn fields. at the University of Illinois.

“Our approach exploits natural variation in genome-wide expression and associated phenotypes within or between species,” added Chia Yi Cheng from the NYU Center for Genomics and Systems Biology and National Taiwan University, the lead author of this study. “We show that reducing our genomic contribution to genes whose expression patterns are conserved within and between species is a biological way to reduce the dimensionality of genomic data, which significantly improves the ability of our machine learning models to identify genes important for a trait. “

Additionally, researchers have proven that this evolutionary-informed machine learning approach can be applied to other traits and species by predicting additional traits in plants, including biomass and yield of Arabidopsis and maize. . They also showed that this approach can predict genes important for drought resistance in another staple crop, rice, as well as the consequences of disease in animals through the study of mouse models.

“Because we have shown that our evolutionary informed pipeline can also be applied to animals, it highlights its potential to discover important genes for any physiological or clinical traits of interest in biology, agriculture or medicine”, Coruzzi said.

In addition to Moose, Coruzzi and Cheng, other researchers involved in this study include co-PIs Ying Li and Kranthi Varala, professors in the Department of Horticulture and Landscape Architecture at Purdue University, as well as members of their research teams at NYU, the University of Illinois, and Purdue. The research was funded by the National Science Foundation’s Plant Genome Research Program (IOS-1339362), the United States Department of Agriculture’s Hatch Project (1013620), the USDA-NIFA Undergraduate Fellowship ( 2016-67011025167) and an NSF CompGen Grant.