DNA Sequence Driven Machine Learning for Modelling Replication Timing - Thesis Overview

Aug 2, 2026

Upload by Aurora

All human somatic cells replicate the genome in S-phase, producing replication timing (RT) profiles that remain consistent across tissues and diseases. The work shows that engineered DNA sequence features are strongly linked to overall RT behaviour and can model aggregate RT profiles from 131 experiments covering 56 human cell types. It further examines how in silico DNA sequence modifications shift predicted RT values. Finally, a single modelling framework is extended to cell-type specific predictions by adding minimal ATAC-seq information capturing chromatin context, enabling both accurate RT prediction and interpretable insight into replication sequence determinants.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages187

Reading Time7 h 51 min

AccessFree to download

Genome Replication Literature

Composition of the Human Genome

Information in the Human Genome

Replicating the Human Genome

Summary

Replication Timing Literature

Experimental Methodologies

Human RT Behaviours

Existing Software

Sequence-Based Modelling

Encode: Informative DNA Encodings

Embed: Information Propagation and Interpretation

Predict: What can be effectively modelled?

Tools for Sequence-Based Modelling

Replication Timing Dataset

Database Sourcing

Database Processing

Consistent RT Behaviour

Average-Behaviour Modelling

Modelling Dataset

Context Window

Feature Extraction

Comparing Model-Type Performance

Optimising Final Model

One-Hot Encoded Deep Modelling

Overall Modelling Decisions

DNA Sequence Driven Machine Learning for Modelling Replication Timing James Ashford Brasenose College University of Oxford A thesis submitted for the degree of Doctor of Philosophy Michaelmas 2023 Live long, and prosper Acknowledgements The last 4 years have been an intense climb, and I owe a lot to the many people who’ve kept me going. First, for the dedication of their time and expertise in both medical sciences and surviving academia, I would like to thank my supervisors Aleksandr Sahakyan and Marella DeBruijn. Their unwavering certainty that I would make it to the top of this cliff-face has been a guiding beacon. I’d also like to thank the MRC for selecting me for the WIMM Prize Studentship that has made my studies in Oxford possible. Secondly, I would like to express my gratitude to the colleagues and new friends I’ve met in my time at Oxford. I’m so glad I got to spend time with my Sahakyan group-mates, especially Patrick and Liezel who I can always count on being in the WIMM for a cup of coffee. Nicole and Tani, you were incredible housemates who brought so many of our cohort together for parties and catch ups, and I look back on our time in Jeune St. fondly. Outside the WIMM, I feel very fortunate to have lived and met people in Brasenose College, especially during my time on the HCR Committee. I’m especially thankful for meeting Ewan, and I hope we will always lose track of time as we walk around Christchurch Meadows sharing our geekiest ideas. I know that the last few years would not have been manageable without help from old friends, in person and through a screen. I’d like to thank the always Eager Beans for the laughs and gossip, and in particular Edmond, Angus, Tom, and Catherine for our regular gaming sessions. I’d also like to thank Bill for his sage DPhil advice, pragmatism, and taste in coffee shops. If love and support became words on the page, my loved ones have written this thesis for me 1000 times over. I’d like to thank my partner Ayesha for her support and understanding during the rollercoaster of emotions that were channeled into finishing the DPhil, and for helping me see it through to the end. My family have been a source of constant support and reassurance, and knowing that I can always turn to you all for advice and comfort makes me one of the luckiest people in the world. I love you all so much, and I hope one day I can help lift you all as high as you have me. James Ashford Brasenose College, Oxford January 9, 2024 Abstract All human somatic cells copy their entire genome during mitotic replication, in the S-phase of the cell cycle. Replication timing (RT) is the temporal order of genome replication in S-phase and has been shown to have consistent global “profiles” across a wide range of tissues and diseases. We demonstrate that while there are many factors that influence the specific RT characteristics of individual cell types, there is a strong link between the DNA sequence composition and the overall RT behaviour. This is achieved by accurately modelling the aggregate profiles from 131 RT experiments constituting 56 unique human cell types, using only engineered features of the DNA sequences as input. We then derive insight into how the composition of DNA sequences impacts RT values, by observing the impact of in silico sequence modifications on model predictions. We further extend our modelling towards cell-type specific predictions with a single model by incorporating a minimal source of extra information, ATAC-seq, which provides context for chromatin organisation. The obtained machine learning models, along with the underlying exploratory data analyses and feature engineering, are both useful for prediction of RT and shed light on the underlying DNA sequence basis of the replication phenomenon. • The behaviour of replication timing (RT) across cell types is analysed by utilising a large cohort of biological samples. • A model of the core, cell type invaria

FAQs

What is replication timing (RT) and why is it important?

RT is the temporal order of genome replication during S-phase. Its profiles show consistent behaviour across tissues and diseases, making it a useful lens for understanding replication regulation.

How does the thesis model RT using only DNA sequence information?

It trains machine learning models on engineered features derived from DNA sequences to reproduce aggregate RT profiles from 131 experiments across 56 human cell types.

How is cell-type specific prediction achieved in the final modelling approach?

A single model framework is extended by incorporating minimal extra information from ATAC-seq, which provides context about chromatin organisation and supports cell-type specific RT modulation predictions.

Aug 2, 2026

Upload by Aurora