Miscanthus AI- Plant selection and breeding for Net Zero

Human selection of crop plants for particular purposes such as food, fibre and fuel has already transformed our world, substantially relieving global hunger during the 20th century but arguably at very significant cost to the global environment from the negative impact of energy inputs and CO2 release from the soil due to changes in agricultural land use. However, intelligent and rapid exploitation of plant diversity, either as crops within intensively managed agriculture or as components of more natural ecosystems, holds promise in addressing NetZero.
Breeding and technology play a major role, minimising inputs such as fertilisers while reducing handling costs in both food and biofuel crops. Genetic changes, whose performance benefits accumulate exponentially across time, represent an excellent investment. Predictive genetic x environment interaction modelling, based on multiple sources of spatio-temporal data (genomics combined with phenomics and environmental information across time and space) holds great promise for accelerated breeding in biofuel crops, many of which have not historically been subjected to selection and breeding.
State-of-the-art plant breeding now interrogates vast quantities of data to understand how the plant genome leads to specific phenotypes or crop traits. However, the application of advanced artificial intelligence to this domain has been relatively unexplored. This project aims to integrate a suite of AI technologies across the plant breeding system, using Miscanthus as the key use case.
We will explore novel machine learning models originating from science discovery within Chemistry to improve genetic prediction and selection for core traits associated with biomass accumulation. These models will be trained with new longitudinal data sets acquired from both the high-throughput phenotyping centre at BBSRC-IBERS (single plants) and with field robots at Lincoln (whole crops). In addition, we will explore how artificial intelligence can augment the decision-making of human plant breeders within the system. Our approach will focus on novel use of computational argumentation to provide an AI-trained logic framework that facilitates explanation-based decision support. This approach has the capacity to not only acquire knowledge over time but also explain decisions to human operators, producing a robotic plant breeder. Our approach bridges the gap between modern genomic selection and human plant breeders.
Data analysis and interpretation by humans and/or autonomous actors is now a bottleneck to exploitation and science discovery. In this project, we will bring together computer scientists, geneticists, and engineers to create an AI-facilitated data analysis pipeline that can rapidly assess, predict and explain plant performance. The outputs will provide a pipeline to accelerate selection of biofuel crops with high yields that are climate resilient and minimise environmental impact.