Project Reference: TRG27-BIO-SB1

Project title: Understanding the function and evolution of transposable elements using AI

Project outline

Transposable elements (TEs) are selfish genomic elements that multiply within the host genome. Although their activity must be tightly controlled to preserve genome integrity, they comprise a vast portion of eukaryotic genomes and are a major driver of both evolutionary innovation and disease. Studying TE distribution and function has long been challenging due to their high sequence diversity and repetitive nature. In this project, you will bridge this gap by training a foundation model using TE sequences from Dfam, an open-access database of repetitive DNA elements. Similar to how large language models learn the grammar and syntax of human language, your model will learn the underlying sequence patterns, diversity, and functional constraints of TEs across the tree of life.

Objectives:

* Develop and pretrain a foundation model for transposable elements.

* Finetune the model for relevant downstream tasks.

* Perform model interpretability analysis to derive new insights into transposon function and distribution.

Your model will be trained using publicly available data in Dfam, which contains 7.6 million transposon families represented as either consensus sequences or hidden Markov models (HMMs). You will first identify a suitable model architecture, data representation and tokenisation strategy, and then pretrain your model using for instance masked language modelling. Once your model is trained, you will fine-tune it across several downstream tasks such as identifying coding sequences and protein domains, classifying novel TEs, predicting host species, and estimating TE age or evolutionary trajectory. Finally, you will analyse the resulting models to reconstruct the rules for TE function and their evolutionary histories.

During this project, you will receive in-depth training in high-performance computing, deep learning, computational biology, comparative and evolutionary genomics, and transposon biology.

Principal Supervisor

Dr Susanne Bornelöv

Email address: smb208@cam.ac.uk

Department / Partner Institute: Biochemistry

Website: https://www.sblab.uk

BBSRC strategic theme

Understanding the rules of life