DrMarina Evangelou
Reader in Statistics
Department of Mathematics - Faculty of Natural Sciences
Orcid identifier0000-0003-0789-8944 (opens in a new tab)
- Reader in StatisticsDepartment of Mathematics - Faculty of Natural Sciences
- 020 7594 7184 (Work)
- 546, Huxley Building, South Kensington Campus, United Kingdom
RESEARCH
Variable and Group Selection
Over the last few years we have developed statistical and machine learning approaches for variable and group selection of high-dimensional data, as the ones seen in genetics and OMICS experiments. For example, we have recently developed a sparse group approach, named sparse group SLOPE, based on the penalised regression approach SLOPE that simultaneously performs variable and group selection while controlling the false discovery rate. For more information see the work of Feser and Evangelou (2023, 2024).
We have alternatively tackled these two challenges through a variational Bayes framework, where a variational Bayes approach for variable selection for survival data analysis has been developed. In addition, a group spike-and-slab variational Bayes approach for different regression models have been developed. For more information see the work of Komodromos, Evangelou, Filippi and Ray (2022, 2023).
Pathway analysis of genome-wide association studies (GWAS) data that aims to select pathways (groups of genes) related to the development of a disease can be considered as a group selection challenge. During my PhD and Postdoc at the University of Cambridge, I worked on the development of Frequentist and Bayesian pathway analysis approaches that utilise either the raw genotype data as well as summary statistics data. For more information see the work of Evangelou, et al. (2012, 2014). One of my most recent work on this topic involves the analysis of pathway-environment interactions for colorectal cancer.
Data Integration
We are further interested in integrating multi-view datasets (datasets from different sources) that describe our samples from different angles. Examples of such datasets include multi-OMICS datasets of the same individuals. We are working on the development of approaches that perform data visualisation of these samples while integrating the multiple data-views, clustering, biclustering and classification algorithms of the multi-view datasets. Such approaches can help towards the identification of different disease subtypes and understand better disease progression. For more information see the work of Rodosthenous, Shahrezaei, Evangelou (2020, 2021, 2024).
Understanding Disease
A lot of my work has been on understanding disease, and on the analysis of different genetics and OMICS datasets. Recent examples of such work include the mendelian randomisation studies between ADHD symptoms and obesity related traits (Karhunen et al. 2021), the associations of miRNA-related sequence variants with clinical diagnoses (Mustafa et al. 2023).
Statistical Cyber-security
During my postdoc at Imperial College London, I worked with Professor Niall Adams on the development of statistical anomaly detection techniques for cyber-security data, and most specifically for enterprise devices (Evangelou and Adams, 2020). Together we co-supervised projects on anomaly detection techniques and on the integration of multiple cyber-security data-sources for clustering. Examples of this work include Workman, Evangelou, Adams (2018, 2021). For a general overview of the statistical cyber-security field have a look at the review article by Passino et al. (2023).
Over the last few years we have developed statistical and machine learning approaches for variable and group selection of high-dimensional data, as the ones seen in genetics and OMICS experiments. For example, we have recently developed a sparse group approach, named sparse group SLOPE, based on the penalised regression approach SLOPE that simultaneously performs variable and group selection while controlling the false discovery rate. For more information see the work of Feser and Evangelou (2023, 2024).
We have alternatively tackled these two challenges through a variational Bayes framework, where a variational Bayes approach for variable selection for survival data analysis has been developed. In addition, a group spike-and-slab variational Bayes approach for different regression models have been developed. For more information see the work of Komodromos, Evangelou, Filippi and Ray (2022, 2023).
Pathway analysis of genome-wide association studies (GWAS) data that aims to select pathways (groups of genes) related to the development of a disease can be considered as a group selection challenge. During my PhD and Postdoc at the University of Cambridge, I worked on the development of Frequentist and Bayesian pathway analysis approaches that utilise either the raw genotype data as well as summary statistics data. For more information see the work of Evangelou, et al. (2012, 2014). One of my most recent work on this topic involves the analysis of pathway-environment interactions for colorectal cancer.
Data Integration
We are further interested in integrating multi-view datasets (datasets from different sources) that describe our samples from different angles. Examples of such datasets include multi-OMICS datasets of the same individuals. We are working on the development of approaches that perform data visualisation of these samples while integrating the multiple data-views, clustering, biclustering and classification algorithms of the multi-view datasets. Such approaches can help towards the identification of different disease subtypes and understand better disease progression. For more information see the work of Rodosthenous, Shahrezaei, Evangelou (2020, 2021, 2024).
Understanding Disease
A lot of my work has been on understanding disease, and on the analysis of different genetics and OMICS datasets. Recent examples of such work include the mendelian randomisation studies between ADHD symptoms and obesity related traits (Karhunen et al. 2021), the associations of miRNA-related sequence variants with clinical diagnoses (Mustafa et al. 2023).
Statistical Cyber-security
During my postdoc at Imperial College London, I worked with Professor Niall Adams on the development of statistical anomaly detection techniques for cyber-security data, and most specifically for enterprise devices (Evangelou and Adams, 2020). Together we co-supervised projects on anomaly detection techniques and on the integration of multiple cyber-security data-sources for clustering. Examples of this work include Workman, Evangelou, Adams (2018, 2021). For a general overview of the statistical cyber-security field have a look at the review article by Passino et al. (2023).