Dr Alexandra Lewin
My research lies at the interface between statistics and machine learning, developing new statistical methods for large-scale data analysis, always with software implementation. Most of my work is on highly-structured, high-dimensional Bayesian models for statistical genomics and genetic epidemiology.
My background is a Maths degree from Cambridge University and PhD in Cosmology from Imperial College London. I taught myself Bayesian statistics during my PhD, and moved into Biostatistics straight after, working briefly on spatial data analysis in the Small Area Health Statistics Unit at Imperial College London, followed by several years on new statistical methods for gene expression data and other 'omics' data in the Department of Epidemiology and Biostatistics at Imperial and in the Maths department at Brunel University. I joined LSHTM in 2018, to work with people across the School on methods and applications using 'omics' data in epidemiology.
I am the co-organiser of the Bayesian teaching on the MSc in Medical Statistics (sub-module Introduction to Bayes in Term 1 and module Bayesian Statistics in Term 2).
I am the co-organiser of the Machine Learning module on the new MSc in Health Data Science, currently under development.
I have also lectured on methods for Big Data in the Advanced Research Methods module on the MSc in Medical Statistics, and contributed to teaching on other modules including the Analysis of Hierarchical and Other Dependent Data.
My current research focus is on Bayesian structural equation models for life-course data from epidemiological cohorts, combining genetic, epigenetic, metabolomic and clinical data. We are developing longitudinal models in which we can induce sparsity over the set of associations amongst the huge set of biomarker/environmental and clinical variables, in order to explore the space of possible mechanistic pathways of disease or disease-related outcomes.
This work is a development of earlier work on Bayesian multivariate regression models, incorporating variable selection (feature selection) and covariance selection.
Previously I worked on Bayesian regression and mixture models for differential gene expression.
I also have an interest in theoretical developments in multiple testing issues in classical methods applied in statistical genetics and 'omics' areas.