Data science

Spring term 2022

24th February 2022 - "Prevalent Cohort Studies: Length-Biased Sampling with Right Censoring" - Masoud Asgharian, McGill University, Canada.

Autumn term 2021

9th December 2021 - "Optimising the Beer Distribution Game: a Reinforcement Learning and Monte Carlo Tree Search Approach" - Felipe Maldonado, University of Essex.

2nd December 2021 - "Sequential Estimation for Mixture of Regression Models in Dealing with Heterogeneous Population" - Hongsheng Dai, University of Essex.

25th November 2021 - "Modelling and Analysis of Individual Animal Movement" - Joe Bailey, University of Essex.

11th November 2021 - "Radial Mendelian randomization: Novel approaches to estimating and visualising causal effects using GWAS summary data" - Wes Spiller, University of Bristol.

4th November 2021 - "Analysing the behaviour of dairy cows" - Kareemah Chopra, University of Essex.

Slope-Hunter: A robust method for collider bias correction in conditional genome-wide association studies

Our department's own Dr Osama Mahmoud led this seminar on bias correction in genetic studies.

Osama explained that studying genetic associations conditioned on another phenotype, for example a study of blood pressure conditional on weight, could be affected by selection bias. An example of this is the study of genetic associations with prognosis (e.g. survival, subsequent events).

Selection on disease status can induce associations between causes of incidence with prognosis, potentially leading to selection bias - also called "index event bias" or "collider bias". At moment one method for adjusting genetic associations for this bias assumes there is no genetic correlation between incidence and prognosis, which may not be a plausible assumption.

Osama proposed the ‘Slope-Hunter’ approach, which has two stages. In the first stage he showed how to use cluster-based techniques to identify: variants affecting neither incidence nor prognosis (these should not suffer bias and only a random sub-sample of them are retained in the analysis); variants affecting prognosis only (excluded from the analysis).

In the second stage, Osama demonstrated cluster-based model to identify the class of variants only affecting incidence. This class was used to estimate the adjustment factor. Simulation studies showed that the approach eliminates the bias and outperforms alternatives in the presence of genetic correlation, and performs as well as alternatives under no genetic correlation when its assumptions are satisfied.

Recent papers

• Elaine Fuertes, Iana Markevych, Richard Thomas, Andy Boyd, Raquel Granell, Osama Mahmoud, Joachim Heinrich, Judith Garcia-Aymerich, Célina Roda, John Henderson, Debbie Jarvis, "Residential greenspace and lung function up to 24 years of age: The ALSPAC birth cohort", Environment International, Volume 140, 2020, 105749, ISSN 0160-4120.
• Roda, Célina and Mahmoud, Osama and Peralta, Gabriela P and Fuertes, Elaine and Granell, Raquel and Serra, Ignasi and Henderson, John and Jarvis, Deborah and Garcia-Aymerich, Judith (2020) 'Physical-activity trajectories during childhood and lung function at 15 years: findings from the ALSPAC cohort.' International Journal of Epidemiology, 49 (1). 131 - 141. ISSN 0300-5771
• Talaei, Mohammad and Hughes, David A and Mahmoud, Osama and Emmett, Pauline M and Granell, Raquel and Guerra, Stefano and Shaheen, Seif O (2021) 'Dietary intake of vitamin A, lung function, and incident asthma in childhood.' European Respiratory Journal. ISSN 0903-1936

Paving the Road to Open Science via Automated Data Documentation

In this talk University of Essex graduate Ahmed Abdelmaksoud looked at how research can be transformed by better data documentation.

Many research projects collect data, which gives us an unparalleled opportunity to not only produce integrated research, but also improve open-access. For example, socio-economic projects could collect and generate data that would also be useful for a health research project. This pre-collected data could help shorten the research timeframe for the health project, but any insights from a health perspective could be used to further generate impact on the original socio-economic project, or provide a foundation for a third piece of research.

Additionally, better data collection and documentation of the process of collecting the data can allow other researchers to replicate a study, proving that the conclusions drawn in the original are accurate (or alternatively identifying issues that need to be addressed).

At the moment there is no standardised way to document data, especially as it often has to be done manually and requires a great deal of time and effort for little reward beyond altruism.

In his talk Ahmed discussed a potential automated solution. Called ‘D2’, it is written in ‘R’ and mainly employs ‘R-Markdown’ and ‘Knitr` to generate documentation for the most commonly used data formats. Minimal input is required from the user, so researchers can utilise it without it taking up too much time.

Ahmed's proposal was particularly interesting as the pandemic and rush to create a Covid-19 vaccine have shown the importance of open science and sharing of data. A tool like this could be a cost and time-effective way to improve this in the future.

Selection bias, missing data and causal inference

In this talk Professor Kate Tilling from the University of Bristol discussed her work on data and bias.

When utilising data different statistical methods can be used for causal inference, but this involves using untestable assumptions. Kate used directed acyclic graphs (DAGs) to show that causal inference is also impacted by selection (for example picking people to take part in a study) or missing data, and gave several examples in which these issues cause bias within studies.

It was a good reminder for our data science audience to keep bias in mind when working with data, particularly when it comes to working in the fields of health and medicine.

Related papers

• Fernando Pires Hartwig, Kate Tilling, George Davey Smith, Deborah A Lawlor, Maria Carolina Borges, "Bias in two-sample Mendelian randomization when using heritable covariable-adjusted summary associations", International Journal of Epidemiology, 2021.
• Fernández-Sanlés A, Smith D, Clayton GL et al. "Bias from questionnaire invitation and response in COVID-19 research: an example using ALSPAC" [version 1; peer review: 1 approved]. Wellcome Open Res 2021, 6:184.

Approximating images by optimally arranging polygons: a heuristic study into computational art

Dr Daan van den Berg, from the University of Amsterdam, gave a talk on his research on computational art.

Daan started off explaining that it is possible to approximate artistic images from a limited number of stacked semi-transparent coloured polygons. He demonstrated with with sets of randomised polygons and showed how to find optimal arrangements for several well-known paintings using three iterative optimisation algorithms: stochastic hillclimbing, simulated annealing and the plant propagation algorithm.

A discussion was then held on the performance of the algorithms, and Daan demonstrated how the found objective values related to the polygonal invariants.

Related papers

• R Dahmani, S Boogmans, A Meijs, D Van den Berg, "Paintings-from-Polygons: Simulated Annealing", Conference on Computational Creativity (ICCC'20), 2020.

EpiViz: an implementation of Circos plots for epidemiologists

Matt Lee, a PhD student from the University of Bristol, delivered a talk on the use of Circos plots in epidemiology.

Biological pathways involve numerous processes, but epidemiology studies predominantly focus on single exposure and single outcome associations. This is primarily because identifying meaningful intermediate associations that can be taken forward for further analysis is complex.

In his talk, Matt discussed how tools like EpiViz can be used to produce simple and efficient Circos plots for those new to programming and data visualisation. By giving people a tool that makes data visualisation easier to produce, epidemiologists can gain a better understanding of the results of complex epidemiological studies. Greater insight in to the results can help increase the impact of such studies.

Related papers

• Matthew A Lee, George McMahon, Ville Karhunen, Kaitlin H Wade, Laura J Corbin, David A Hughes, George Davey Smith, Debbie A Lawlor, Marjo-Riitta Jarvelin, Nicholas J Timpson, Common variation at 16p11.2 is associated with glycosuria in pregnancy: findings from a genome-wide association study in European women, Human Molecular Genetics, Volume 29, Issue 12, 15 June 2020, Pages 2098–2106

A Statistician’s Botanical Garden - The Ideas behind Trees, Model-Based Trees and Random Forests

Classification and regression trees, model-based trees and random forests are powerful statistical methods from the field of machine learning. However, while individual trees are easy to interpret, random forests are "black box" prediction methods. Despite this, they provide variable importance measures, that are being used to judge the relevance of the individual predictor variables.

In this seminar, Professor Carolin Strobl introduced the rationale behind trees, model-based trees and random forests, and illustrated their potential for high-dimensional data exploration, while also pointing out limitations and potential pitfalls in their practical application.

Related papers

• Fokkema, M., & Strobl, C. (2020). Fitting prediction rule ensembles to psychological research data: An introduction and tutorial. Psychological Methods, 25(5), 636–652.

Detecting the hierarchical structure of the cell nucleus

Chromatin consists of DNA wrapped around histones and forms complex three-dimensional structures within the cell nucleus with various degrees of compaction.

Genes have been shown to be repressed by their proximity to the nuclear periphery or activated by being in contact with special regulatory regions called enhancers. Thus the relative positioning of genes and their interactions with other regions are very important in determining whether they are expressed or not.

In this talk, Iona Olan from the University of Cambridge discussed her work on cellular senescence, a phenotype associated with dramatic changes in its chromatin interactions network relative to normal cells. Senescence corresponds to permanent cell cycle arrest and has been shown to act as a protective barrier against tumourigenesis.

Related papers

• Kosuke Tomimatsu, Dóra Bihary, Ioana Olan, Aled Parry, Stefan Schoenfelder, Adelyne Chan, Guy Slater, Yoko Ito, Peter Rugg-Gunn, Kristina Kirschner, Camino Bermejo-Rodriguez, Masako Narita, Tomomi Seko, Hiroyuki Kugoh, Ken Shiraishi, Koji Sayama, Hiroshi Kimura, Peter Fraser, Shamith Samarajiwa, Masashi Narita, Locus-specific induction of gene expression from heterochromatin loci during cellular senescence, Nature Research, pre-print.