Project Reference: TRG27-PAT-MB1

Project title: Developing large language models to understand how DNA replication is shaped by evolution

Project outline

DNA replication is one of the most fundamental processes in cell biology, yet we still understand remarkably little about where it begins. Replication starts at discrete genomic sites called origins of replication, but in humans and most organisms, what determines whether a genomic locus can act as an origin has remained largely unknown. Until recently, the prevailing view was that primary DNA sequence does not govern whether a particular locus acts as a replication origin in the human genome. We overturned this view by developing ORILINX, a large language model that predicts replication origins from primary DNA sequence alone. Although ORILINX was trained on human data, it generalises across mammals (mouse, sheep) and even birds (chicken). This project will explore how and why origin locations change over evolutionary time.

The student will use ORILINX to call replication origins genome-wide across mammalian species spanning major clades (primates, rodents, and laboratory and livestock species). From these maps, they will build phylogenetic trees describing how organisms replicate their DNA and compare these against established sequence-based phylogenies. There will be ample opportunity to build upon, further develop, and expand ORILINX so that it generalises across an increasingly wide range of organisms.

Skills gained through this project will include applying AI models to biological data, high-performance computing, familiarity with bioinformatics file formats, computer programming, managing large datasets, and leading an interdisciplinary research project. Our team of computational scientists and software developers has extensive experience providing supportive mentorship. This project is therefore well-suited to individuals who are new to programming and computational biology but eager to take their first steps in using and building AI models; it can also be made more advanced for students with prior computational biology experience.

Principal Supervisor

Dr Michael Boemo

Email address: mb915@cam.ac.uk

Department / Partner Institute: Pathology

Website: https://www.boemogroup.org/

BBSRC strategic theme

Understanding the rules of life