Analysis of gene regulation using sequencing data. We are developing the EpiSignal pipeline, which uses genomic annotation to interrogate sequence data and identify potentially differentially regulate regions from different modaliteis: genomic expression (mRNA-seq), DNA accessibility (ATAC-seq), and histone modification (ChIP-seq) data. Resulting regions are categorised into on (or signal) and off (or noise) states by an EM algorithm that assumes a mixture of two normal distributions. While data from within a cell type show mainly quantitative differences between genotypes, differences between cell types are dominated by a distinctive qualitative on/off pattern. 

Genome Segmentation Algorithms. We recently developed oHMMed, a class of ordered HMMs with emission densities, to identify comparably homogeneous regions within autocorrelated genomic sequences. HMMs are classically used to segment genomic sequences, oHMMed additionally accounts for autocorrelations by assuming that emission densities are ordered by a key parameter and only allows transitions between states that are neighbours in this ordering. We are working on generalising these models by extending them to other emission density families and data classes and by removing the current restriction to constant input window sizes. 

Estimation of population genetic forces from site frequency data. We develop probabilistic models and statistical approaches to analyse population genetic data, predominantly to infer population genetic forces from site frequency spectra. The modelling approaches always arise from population genetic theory, often rooted in Moran or corresponding diffusion models. Specifically, we focus on deriving closed form maximum likelihood estimators from (bi-allelic) frequency spectra. The aim is to develop rigorous statistical methods to efficiently analyse currently available population genetic data. 

Main Techniques 

Probabilistic modelling, statistical theory, scripting in the statistical programming language “R”