Trees and Forests: Data Science and Machine Learning Approaches to Characterise Clonal Haematopoiesis

Clonal haematopoiesis (CH), the expansion of a haematopoietic stem cell (HSC) driven by somatic mutations in leukaemia-associated genes, is a common age-related phenomenon that affects more than 20% of adults aged over 70 years. As the shared precursor of most myeloid neoplasms (MN), timely detection of CH can offer an opportunity for cancer prevention through the interception of premalignant clones. Achieving this goal requires that several key challenges are overcome, including: the scalable identification of individuals with high-risk clones, the accurate stratification of CH subtypes most strongly linked to progression, and the elucidation of the mechanisms by which driver mutations confer a fitness advantage to guide therapeutic intervention. In this thesis, I described our attempts to address these challenges through a variety of distinct but related approaches. Firstly, I describe the development of CHIC (Clonal Haematopoiesis Inference from Counts), a framework that uses tree-based machine learning classifiers to predict the presence of CH from complete blood count indices. Next, I examine the prevalence of Clonal Monocytosis of Undetermined Significance (CMUS) in the UK Biobank, describe its association with haematological and non-haematological diseases, and propose refinements to its definition that strengthen its clinical relevance and association with MN. Subsequently, I focus on CH driven by splicing factor mutations, which impart a fitness advantage to HSCs via an unknown mechanism. Through phylogenetic reconstruction of haematopoietic cell colonies, I demonstrate that, compared to colonies without CH mutations, colonies bearing common driver mutations exhibit shorter telomeres consistent with clonal expansion, whereas those with splicing factor mutations paradoxically display longer telomeres. This suggests that splicing gene mutations rescue HSCs from critical telomere shortening, thus restoring or augmenting their clonal fitness. Finally, I report and characterise the first non-coding drivers of sporadic CH, namely mutations in the TERT promoter, outlining how they mirror the age distribution of splicing factor mutations and portend a risk of pulmonary fibrosis. Taken together, this body of translational work develops novel strategies for scalable CH detection, establishes data-driven refinements to diagnostic classification of clonal monocytoses, and sheds light on the selection pressures driving the emergence of high-risk splicing factor mutant clones.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2025

Author dc:creator

• Dunn, William Grant

Advisors dc:contributor.advisor

• Vassiliou, George
• Mohorianu, Irina

• Clonal haematopoiesis

Licence

• https://www.repository.cam.ac.uk/bitstreams/1c1f5107-fe9e-438c-9c48-9311281c28d4/download
• http://purl.org/NET/rdflicense/allrightsreserved

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.126844

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/397815