A recent article by the Crop Science Society of America titled “Deep Learning: The Future of Genomics-Driven Plant Breeding?” by DJ McCauley summarizes the importance of machine learning and its role in crop improvement.
The article states, “Machine Learning is a new tool to help with predictions in plant and animal breeding - one that looks to the future, to the growing omics disciplines for deeper structure, statistical backing, and streamlined selection.” And, quoting José Crossa, study author, “Machine learning will not make the skillsets of plant breeders obsolete. Machine learning maximizes the role of plant breeders.”
At Computomics, we fully agree with this assessment. Our disruptive machine learning technologies help breeders design better breeding programs. Computomics' technology predicts phenotypes such as yield and plant height for the next generation of offspring from the genetic information of the parental crops.
Recent research has clearly shown that weather, soil and other environmental conditions can also impact plant yield. The interaction between a plant and its environment can be very complex and each environment can effect a plant differently.
As this article correctly points out, “What’s fantastic about deep learning is the flexibility it offers to produce predictions for multiple traits with different kinds of responses and for multi-environment data”. Our machine learning technology combines the genetic information of plants with phenotyping information and meta data to identify advantageous locations around the world. It takes into account past, present and future weather patterns, pollution levels and soil information to help our customers discover new and avoid redundant locations.
As a part of our CropML project, we are developing a pipeline that uses multiple machine learning models such as Random Forest, Extreme Gradient Boosting and Deep Neural Networks to study the interaction of the crop with its environment, to predict yield and other traits. We are also developing data analysis methods that check the quality of the environmental data, allowing customers make better decisions on where and when to plant their crops.
We have successfully implemented our pipeline on soy line and hybrid corn breeding programs and shown that combining data from multi-environment trials boosts the prediction accuracy while saving time and resources.
Dr. Rupashree Dass
Machine Learning Scientist
Find out more about the applications of machine learning in the article above. Contact Rupa to learn more about how Computomics' machine learning technology can support your breeding program!