phd · warwick medical school · 2026
Reading immune cells from the DNA they keep open.
I build machine-learning methods that recognise immune cell types from single-cell chromatin accessibility data, and I test how well they hold up when they move from blood into tumours.
An illustration of ATAC-seq, the assay behind my research. Not real data.
I came to biology through computing. After a BSc in Computer Engineering at Kocaeli University in Turkey and early work as a software developer, I took an MSc in Computer Science at the University of Nottingham, with a dissertation on algorithmic puzzles in Haskell.
For my PhD I joined Professor Sascha Ott’s group at Warwick Medical School, with Professor Lawrence Young as second supervisor. My research asked whether we can tell which immune cell we are looking at from the parts of its DNA that are open. I built a classifier to do this and tested how far it carries from blood into tumours.
My classifier was built to stop at a broader label when the evidence is weak, rather than guess. I learn most by studying where a model goes wrong.
My work sits where machine learning meets single-cell genomics and immunology.
Single-cell ATAC-seq shows which stretches of DNA are open in each individual cell. These open regions differ between cell types, and they can keep a record of a cell’s past activity. That makes them useful fingerprints for telling immune cells apart, even without measuring RNA.
Blood cells develop along a branching path from stem cells. I arrange classification the same way, with a small support vector machine for each box in the tree, so each cell is labelled step by step from “immune” down to a specific subtype. When the evidence is weak, a cell often stops at a broader label rather than being forced into a narrow one.
A model trained on blood does not automatically work in a tumour. Some of its marker regions are missing from tumour data, and freezing, sample handling and genome versions all change what the model sees. I measured which markers survive the move. Mature immune cells kept their marker regions much better than early progenitor cells did.
Knowing a cell’s type is the first step. The next is its state, for example whether a T cell inside a tumour is resting, active or exhausted, and whether that pattern relates to how patients respond to treatment. I am also interested in combining chromatin data with other measurements taken from the same cells.
phd thesis · warwick medical school
Thesis record on WRAP Full text available from June 2028.
Almost every cell in the body carries the same DNA, yet an immune cell behaves very differently from a tumour cell. Part of the reason is packaging. Some stretches of DNA are wound tightly away, while others stay open and ready to be read. My thesis built a method that reads these open regions in thousands of single cells and decides which immune cell each one is. It asks a series of yes-or-no questions that follow the family tree of blood cells.
I trained it on a public atlas of human blood and bone marrow, then tested it on tumours. When it gave a label, the label was usually right. Inside tumours, mature immune cells kept most of their marker regions, while early-stage cells lost many. Keeping the markers was not always enough. Closely related subtypes, such as naive and memory T cells, stayed hard to tell apart.
Each box is a yes-or-no classifier. A cell enters at the top and moves down only while each classifier agrees.
Cells sent through here are illustrative, not results from the thesis.
Real blood development includes cells caught between stages. These transitional cells were the hardest for the classifier to place.
2020 – 2026
Thesis: Epigenome-informed Computational Cell Sorting. Ott group, supervised by Professor Sascha Ott and Professor Lawrence Young. Funded by an overseas postgraduate scholarship from the Republic of Turkey Ministry of National Education.
2026
Measured, cell by cell, how much RNA Polymerase II sits in the nucleus versus the cytoplasm in confocal microscopy images. Built in Python with Cellpose and scikit-image.
2017
Dissertation on algorithmic puzzles in Haskell, awarded a distinction. Focus on advanced algorithms, functional programming and machine learning.
2012
Built a customer management module in ASP.NET and SQL Server, and worked in a team of five on client-specific features.
2012
2011
Built GIS data visualisation tools in C for the urban planning department.
2010
Built a web-based compliance tracking system in JSP and MySQL.
2024 · Poster
“Epigenome Informed Cell Sorting of Immune Cells”
Warwick-Cambridge Quantitative Cell Biology Symposium, University of Warwick
2023 · Talk
“Epigenetic Codes of Immunosuppression”
Warwick Medical School Postgraduate Research Symposium
2022 · Poster
“Epigenetic Codes of Immunosuppression”
Gordon Research Conference on Single-Cell Cancer Biology, Easton, Massachusetts, USA
Supported by Cancer Research UK and a Warwick Medical School travel grant.
2022 · Poster
“Epigenetic Codes of Immunosuppression”
Warwick Medical School Postgraduate Research Symposium
I am glad to hear from people working on single-cell genomics, immunology or machine learning, whether about research or roles.