An article recently published by the InnovPlantProtect (InPP) team reveals the potential of computational machine learning methods for predicting phenotypic traits, such as wheat yield, from genetic information of this plant.
Machine learning(ML) is an area of data science that has gained increasing relevance in the last decade. ML is a branch of artificial intelligence that allows the development of predictive models that can be applied in a wide range of areas. Although we don’t realize it, we use ML-based tools in our day-to-day, such as personalized results displayed in your Facebook feed. But future applications range from enabling autonomous driving to detecting diseases through the analysis of radiographs (in humans) or drone images (in orchards).
Genomic prediction (GP) is another area in which ML has been applied. This involves using genomic data (which give us information about the genotype) to develop computer models for predicting complex phenotypic traits of organisms, such as wheat yield (See schematic representation).
In this research, now published in the scientific journal Agriculture, the researchers Manisha Sirsat and Ricardo Ramiro, both from the Data Management and Risk Analysis department, in collaboration with Paula Oblessuc from the Protection of Specific Crops department, explored the use of several genomic prediction (GP) models based on different computational methods in addition to ML, such as statistical or deep learning (DL) methods, with the objective of comparing the robustness and performance of each one of them in predicting the phenotypic trait of wheat yield. The idea was to understand which methods allow predicting phenotypic traits with greater reliability.
“Statistical methods have been the most commonly used in GP by research teams worldwide. However, ML methods are proving to be a good alternative for GP in terms of accuracy, computational time, and cost”, highlights Manisha Sirsat, the first author of the study.
“GP based on ML can help in reducing the time and cost of extensive phenotyping evaluation (during breeding programs) and in accelerating the genetic gain”, explains the researcher. “This study contributes to helping researchers understand the key factors in the development of models that can accelerate wheat breeding programs, or other crops, and increase agricultural productivity”, she adds.
The team has been working on genomic prediction since 2020, and it expects that Genomics and GP are central in allowing crop yield to be maintained/increased, despite the multiple threats we face, and to respond to the 50% increase in food demand by 2050, as the global population reaches 9,7 billion.
Researchers Manisha Sirsat, Ricardo Ramiro e Paula Oblessuc (from the left to the right)
Manisha Sirsat e Ricardo Ramiro