Leveraging epigenetic signatures to predict cell-type of origin from nanopore sequencing | LC26

Abstract
DNA methylation varies across tissues and cell types, which complicates the interpretation of disease-associated epigenetic signatures in mixed tissues such as whole blood. Generating cell-type-specific methylation profiles is therefore essential, but physical isolation of cell subtypes is labour intensive and not feasible for large studies. Long-read sequencing offers an opportunity to infer cell identity directly from read-level methylation patterns. We aimed to assess whether long-read methylation data can be used to classify sequencing reads by blood cell type and to evaluate the feasibility of creating cell-type-resolved methylation profiles computationally. Reference DNA methylation data for six common blood cell types were used to train models for read-level cell-type prediction. The genome was partitioned into windows comparable to long-read lengths to evaluate classification performance across the genome and across and cell types. Cell type could be accurately predicted from relatively small genomic regions, although performance varied across the genome and between cell types. Approximately one third of the genome enabled reliable discrimination between lymphoid and myeloid lineages. Several regions also supported accurate classification of more specialised immune subtypes. Conclusions: Long-read sequencing data contain sufficient methylation information to infer cell identity without physical cell sorting. This computational approach offers a scalable alternative for generating cell-type-specific methylation profiles in epigenetic epidemiology. Further work will expand reference data, refine classifiers, and test performance in larger datasets.

Biography
Eilis Hannon is an Associate Professor in Bioinformatics at the University of Exeter Medical School and an Engineering and Physical Sciences Research Council (EPSRC) Research Software Engineering Fellow. Her research focuses on integrating genetic, transcriptomic and epigenomic data to understand molecular mechanisms underlying brain disorders. With expertise in statistical genomics, she develops computational methods to study cell-type-specific epigenetic variation. Eilis also leads research on data science pedagogy through her role as director of the Coding for Reproducible Research programme and a Turing Institute Skills Policy Award project.