Understanding the evolution of crops using deep learning and evolutionary models

Brassicas are the crop species which are most closely related to the well-studied model plant
Arabidopsis. Interestingly, they often have multiple copies of Arabidopsis genes, due to polyploidisation
(whole genome duplication). This project will investigate how the regulatory DNA sequences of
Brassicas have evolved, using methods that will allow quantitative comparison with alternative
evolutionary models.
Artificial intelligence algorithms that are used to annotate texts and images have recently been applied
to interpret DNA sequences. The Ezer lab has developed an AI for annotating new regulatory regions in
Arabidopsis, based on DNA sequence (unpublished preliminary results acquired through a Turing
Institute research grant). This data is trained on large databases of pre-existing RNA-seq data.
The PhD student will refine this AI model and then apply it to Brassicas, where extensive RNA-seq
datasets have recently been made available. In parallel, the student will develop and implement various
evolutionary mathematical models of promoter evolution. They will compare the annotated regulatory
sequences with the predictions made by the evolutionary models. This will help us identify which
evolutionary model best explains the evolution of regulatory sequences in Brassicas. As most of these
models have been developed in non-plant species and in the absence of polyploidisation, this will shed
an important light on the similarities and differences between regulatory sequence evolution across
biological kingdoms. The work will also show how mathematical models can be used in conjunction
with AI to arrive at a practical understanding of the evolution of crop systems.