Skip to main content Skip to navigation

Forum Talk Abstracts

Wednesday 22 June 2022 -

Wednesday 08 June 2022 -

Wednesday 25 May 2022 -

Wednesday 11 May 2022 - Rebecca Killick (Lancaster)
Tackling Global Issues: Change

Traditional statistical model building assumes that the same model (and fitted parameters) can describe the data at any point in the observed process. To tackle early violations to this statisticians introduced regressors to describe time-varying features such as trend and seasonality. These have enabled model building to become engrained in everyday applications across all fields. But what happens when this neat assumption of a static model is no longer appropriate?

The simplest departure from this static model assumption is to piece together static models, which we understand well. Questions then surface around where and how many of these static models should we use? This field of statistical modelling is called changepoint detection.

This talk will give a potted history of the challenges in changepoint detection that I have been preoccupied with over the last ten or so years. Always motivated by applications of changepoint detection to real-world problems, the tour of statistical challenges will be littered with applications to different areas including environment, health and business. To conclude the talk I will describe some of the existing challenges that must be addressed by the community if changepoint detection is going to remain relevant to the world.

Within the talk, my aim is to convey the intuition behind the field, its challenges and proposed solutions. As such, the talk will be accessible to a general audience.

Wednesday 16 March 2022 - Nicole Nisbett (Leeds)

Unearthing climate action discourses with sentence transformers

In this talk I explain how we use natural language processing and the BERTopic algorithm to uncover topics from text. The algorithm uses a transformers-based topic modeling approach, embedding models, clustering, and vector space representations to provide meaning to unstructured text data. We apply these models to large volumes of Twitter data on the climate crisis discussion around the UN Climate Change Conferences (COP) to understand the core themes in the public debate and how these have changed between 2014 and 2021.

Wednesday 09 March 2022 - Yonatan Berman (KCL)

A Simplified Mortality Multiplier Method: New Estimates of Wealth Concentration

We derive a method for estimating top wealth shares using estate tax data and minimal data on mortality. It is a simplification of the mortality multiplier method. We first investigate the conditions in which the novel method provides accurate wealth concentration estimates and validate it empirically. We then use it to produce novel long-run historical series of the distribution of wealth for Belgium, Japan, and South Africa, where data have not been exploited yet. This is particularly important for historical inequality studies, given that existing series of wealth inequality are very scarce and only cover a few developed countries.

This is joint work with Facundo Alvaredo and Salvatore Morelli

Wednesday 02 March 2022 -

Wednesday 16 February 2022 - Simon Schnyder (Kyoto)

Epidemics of infectious diseases posing a serious risk to human health have occurred throughout history. During the ongoing SARS-CoV-2 epidemic there has been much debate about policy, including how and when to impose restrictions on behavior. Under such circumstances policymakers must balance a complex spectrum of objectives, suggesting a need for quantitative tools. Here we show how costly interventions, such as taxes or subsidies on behaviour, can be used to exactly align individuals' decision making with government preferences even when these are not aligned. We assume that choices made by individuals give rise to Nash equilibrium behavior. We focus on a situation in which the capacity of the healthcare system to treat patients is limited and identify conditions under which the disease dynamics respect the capacity limit. In particular we find an extremely sharp drop in peak infections as the maximum infection cost in the government's objective function is increased. This is in marked contrast to the gradual reduction without government intervention. The infection costs at which this switch occurs depend on how costly the intervention is to the government. We find optimal interventions that are quite different to the case when interventions are cost-free.

Wednesday 09 February 2022 - Tommaso Lorenzi (Politecnico di Torino)
Dissecting the evolutionary and spatial dynamics of cancer through partial differential equations

A range of mathematical models have been used to gain a more in-depth theoretical understanding of different aspects of cancer dynamics. In this talk, deterministic, continuum models formulated as partial differential equations will be considered. The first part of the talk will focus on partial integro-differential equations modelling the eco-evolutionary dynamics of cancer cells in vascularised tumours. In the second part of the talk, attention will turn to models of avascular tumour growth that comprise coupled systems of nonlinear partial differential equations, which reflect the heterogeneous cellular composition of the tumour micro-environment. Analytical and numerical results summarising the behaviour of the solutions to the model equations will be presented and the biological insight generated by these results will be discussed.

POSTPONED

Wednesday 02 February 2022 - Rebecca Killick (Lancaster)

Tackling Global Issues: Change

Traditional statistical model building assumes that the same model (and fitted parameters) can describe the data at any point in the observed process. To tackle early violations to this statisticians introduced regressors to describe time-varying features such as trend and seasonality. These have enabled model building to become engrained in everyday applications across all fields. But what happens when this neat assumption of a static model is no longer appropriate?

The simplest departure from this static model assumption is to piece together static models, which we understand well. Questions then surface around where and how many of these static models should we use? This field of statistical modelling is called changepoint detection.

This talk will give a potted history of the challenges in changepoint detection that I have been preoccupied with over the last ten or so years. Always motivated by applications of changepoint detection to real-world problems, the tour of statistical challenges will be littered with applications to different areas including environment, health and business. To conclude the talk I will describe some of the existing challenges that must be addressed by the community if changepoint detection is going to remain relevant to the world.

Within the talk, my aim is to convey the intuition behind the field, its challenges and proposed solutions. As such, the talk will be accessible to a general audience.

Wednesday 19 January 2022 - Diana Khoromskaia (Crick Institute)
Active morphogenesis of patterned epithelial shells
Shape transformations of epithelial tissues in three dimensions, which are crucial for embryonic development or in vitro organoid growth, can result from active forces generated within the cytoskeleton of the epithelial cells. How the interplay of local differential tensions with tissue geometry and with external forces results in tissue-scale morphogenesis remains an open question. Here, we describe epithelial sheets as active viscoelastic surfaces and study their deformation under patterned internal tensions and bending moments. In addition to isotropic effects, we take into account nematic alignment in the plane of the tissue, which gives rise to shape-dependent, anisotropic active tensions and bending moments. We present phase diagrams of the mechanical equilibrium shapes of pre-patterned closed shells and explore their dynamical deformations. Our results show that a combination of nematic alignment and gradients in internal tensions and bending moments is sufficient to reproduce basic building blocks of epithelial morphogenesis, including fold formation, budding, neck formation, flattening, and tubulation.

Preprint: D. Khoromskaia and G. Salbreux, arXiv:2111.12820 [physics.bio-ph]

Wednesday 12 January 2022 - Sophie Meakin (London School of Hygiene and Tropical Medicine, and MathSys alumna)
Comparative assessment of methods for short-term forecasts of COVID-19 admissions in England at the local level
Forecasting healthcare demand is essential in epidemic settings, both to inform situational awareness and facilitate resource planning. We made weekly forecasts of daily COVID-19 hospital admissions for National Health Service (NHS) Trusts in England between August 2020 and April 2021 using three disease-agnostic forecasting models: a mean ensemble of autoregressive time series models, a linear regression model with 7-day-lagged local cases as a predictor, and a scaled convolution of local cases and a delay distribution. We compared their point and probabilistic accuracy to a mean-ensemble of them all, and to a simple baseline model of no change from the last day of admissions. We measured predictive performance using the Weighted Interval Score (WIS) and considered how this changed in different scenarios (the length of the predictive horizon, the date on which the forecast was made, and by location), as well as how much admissions forecasts improved when future cases were known.

Wednesday 08 December 2021 - Rosanna Barnard (London School of Hygiene and Tropical Medicine)
Modelling SARS-CoV-2 transmission in England: vaccines, variants and variability
In this talk I will present mathematical modelling work considering the dynamics of SARS-CoV-2 transmission in England. I will discuss the challenge of modelling disease transmission within an ever-evolving epidemiological context, with consideration of COVID-19 vaccinations, variants of concern, and various uncertainties such as behavioural changes. I will outline the compartmental modelling framework we use to describe SARS-CoV-2 transmission, as well as the data sources and the model fitting processes used. I will give an overview of various modelling outputs we have produced which have informed the UK response.

Wednesday 01 December 2021 - David SibleyLink opens in a new window (Loughborough)
Coupled dynamics of fluid systems with mass transfer
In this talk we will consider how to develop models for systems of fluids that involve complex behaviour such as mass transfer. We will consider multiphase systems of a single component in solid, liquid and vapour form, and systems that involve both fluids and colloids. In both situations the dynamics should account for conserved behaviour such as surface tension driven flow, and non-conserved behaviour due to evaporation/condensation. We will discuss results for the solid-liquid-vapour system in detail. We consider a regime where a droplet of liquid spreads over a solid and in the presence of a vapour, all of the same material (such as ice-water-water vapour). We consider dynamics where vapour can condense to liquid which in turn can freeze to solid throughout a droplet spreading process, and indeed vice-versa. By developing a thin-film model capturing phase-change, surface tension, density contrast, and interfacial potentials, a rich phase diagram will be explored---and complex dynamics such as layered solid growth driven by forces at the contact lines described [work in collaboration with Pablo Llombart (Universidad Complutense Madrid), Eva G. Noya (CSIC Madrid), Andrew J. Archer, and Luis G. MacDowell (Universidad Complutense Madrid); based on Nat. Commun. 12, 239 (2021)].

Wednesday 24 November 2021 - Alejandra Avelos Pacheco (Harvard)
Multi-study Bayesian factor regression analysis for heterogenous high-dimensional biological data
Two key challenges in modern statistical applications are the large amount of information recorded per individual, and that such data are often not collected all at once but in batches. These batch effects can be complex, causing distortions in both mean and variance. We propose a novel sparse latent factor regression model to integrate such heterogeneous data. The model provides a tool for data exploration via dimensionality reduction and sparse low-rank covariance estimation while correcting for a range of batch effects. We study the use of several sparse priors (local and non-local) to learn the dimension of the latent factors. We provide a flexible methodology for sparse factor regression which is not limited to data with batch effects. Our model is fitted in a deterministic fashion by means of an EM algorithm for which we derive closed-form updates, contributing a novel scalable algorithm for non-local priors of interest beyond the immediate scope of this paper. We present several examples, with a focus on bioinformatics applications. Our results show an increase in the accuracy of the dimensionality reduction, with non-local priors substantially improving the reconstruction of factor cardinality. The results of our analyses illustrate how failing to properly account for batch effects can result in unreliable inference. Our model provides a novel approach to latent factor regression that balances sparsity with sensitivity in scenarios both with and without batch effects and is highly computationally efficient; and opens new avenues for future research on dimension-reduction-based data integration.

Wednesday 17 November 2021 - Katerina Kaouri (Cardiff)
Modelling across scales: from fertilization and embryogenesis to COVID-19 modelling
I will present an overview of some interdisciplinary challenges I have been working on, in collaboration with experimentalists and industry. I will start with the important societal challenge of improving In-Vitro Fertilization (IVF) therapies, proceed with models of calcium signalling in IVF and in embryogenesis. I will also relay our work on modelling airborne transmission of COVID-19 in indoor spaces which has been developing over the last year in collaboration with Oxford University and the Welsh government.

Wednesday 10 November 2021 - Adam Gosztolai, Ramdya lab, EPFL
A geometric lens to reveal links between structure and dynamics of networks
Describing networks geometrically is a powerful tool to understand the link between structure and signal propagation as well as in the design of efficient learning algorithms. Traditionally, geometric descriptions involved embedding nodes in low-dimensional manifolds or normed vector spaces. However, an incompatible embedding space can lead to detrimental information loss. In this talk, I will generalise the Ollivier-Ricci edge curvature by quantifying the interaction of a pair of dynamical processes. Crucially, this definition does not require embedding and induces geometric representations at different resolutions through the time parameter of the dynamics, which can unfold the multiscale structure of arbitrary networks. To show this, I will illustrate that the edge curvature unveils deep connections between network structure and signal propagation. Specifically, a single quantity, the edge curvature, bounds the rate of information flow across the edge, revealing network bottlenecks. To capture these bottlenecks, I will introduce a geometric generalisation of network modularity, a quality function that encodes clusters based on the curvature weighted graph without requiring a statistical null model. I will then prove that for sparse stochastic block model networks geometric modularity is optimal in the sense that it recovers clusters until the well-known phase transition of detectability. Finally, I will demonstrate how geometric modularity is useful to capture clusters based on dynamic interactions in real-world networks, including the European power grid and a dataset of C. elegans single neurons gene expressions.
 
Reference: Gosztolai, A., & Arnaudon, A. (2021). Unfolding the multiscale structure of networks with dynamical Ollivier-Ricci curvature. Nature Communications, 12(1), 1–11. https://doi.org/10.1038/s41467-021-24884-1Link opens in a new window

Wednesday 03 November 2021 - Ruth Baker, University of Oxford
Teasing apart the impact of electric fields on single-cell motility
Cell motility in response to environmental cues forms the basis of many developmental processes in multicellular organisms. One such environmental cue is an electric field, which induces a form of motility known as electrotaxis. Electrotaxis has evolved in a number of cell types to guide wound healing, and has been associated with different cellular processes, suggesting that observed electrotactic behaviour is likely a combination of multiple distinct effects arising from the presence of an electric field. In order to determine the different mechanisms by which observed electrotactic behaviours emerge, and thus to design electric fields that can be applied to direct and control electrotaxis, we require accurate quantitative predictions of cellular responses to externally-applied fields. Here, we use mathematical modelling to formulate and parametrise a variety of hypothetical descriptions of how cell motility may change in response to an electric field. We calibrate our model to observed data using synthetic likelihoods and Bayesian sequential learning techniques, and demonstrate that electric fields impact cellular motility in three distinct ways.

Wednesday 20 October 2021 - Gabriela Gomes, University of Strathclyde
Frailty variation in population dynamics: Adventures and misadventures of an elusive concept
Selection acting on unmeasured individual variation is a common source of bias in the analysis of populations. It has been shown to affect measured rates of mortality1-3, the survival of endangered species4,5, the scope of neutral theories of biodiversity and molecular evolution6,7, the measured risk of diseases whether non-communicable8,9 or infections10-15, and the efficacy of interventions such as vaccines16-20 or symbionts21,22. Building on this knowledge, we have addressed how selection on individual variation might have affected the course of the coronavirus disease (COVID-19) pandemic23.

This form of variation that responds to selection and impacts within-cohort population dynamics, termed frailty variation by Vaupel et al. (1979)2, constitutes a most genuine phenomenon that scientific disciplines have been dismissing for decades. I will present some examples and discuss the mixed attitudes towards what is arguably the most elusive concept in population dynamics.

References:

1. Keyfitz, N. & Littman, G. (1979) Popul. Stud. 33, 333-342.
2. Vaupel, J., Manton, K. & Stallard, E. (1979) Demography 16, 439-454.
3. Vaupel, J., & Yashin, A. (1985) Am. Stat. 39, 176-185.
4. Kendall, B. E. & Fox, G. A. (2002) Conserv. Biol. 16, 109-116.
5. Jenouvrier, S, Aubry, L. M., Barbraud, C, Weimerskirch, H & Caswell, H. (2018) J. Anim. Ecol. 87, 212-222.
6. Steiner, U. K. & Tuljapurkar, S. (2012) Proc. Natl. Acad. Sci U. S. A. 109, 4684-4689.
7. Gomes, M. G. M., King, J. G., Nunes, A., Colegrave, N. & Hoffmann, A. (2019) Ecol. Evol. 16, 8995-9004.
8. Aalen, O. O., Valberg, M., Grotmol, T. & Tretli, S. (2015) Int. J. Epidemiol. 4, 1408-1421
9. Stensrud, M. J. & Valberg, M. (2017) Nat. Commun. 8, 1165.
10. Anderson, R. M., Medley, G. F., May, R. M. & Johnson, A. M. (1986) IMA J. Math. Appl. Med. Biol. 3, 229-263.
11. Dwyer, G., Elkinton, J. S. & Buonaccorsi, J. P. (1997) Am. Nat. 150, 685-707.
12. Smith, D. L., Dushoff, J., Snow, R. W. & Hay, S. I. (2005) Nature 438, 492-495.
13. Bellan, S. E., Dushoff, J., Galvani, A. P. & Meyers, L. A. (2015) PLOS Med. 12, e1001801.
14. Gomes, M. G. M., et al. (2019) Nat. Commun. 10, 2480.
15. Corder, R. M., Ferreira, M. U. & Gomes, M. G. M. (2020) PLOS Comput. Biol. 16, e1007377.
16. Halloran, M. E., Longini, I. M. Jr. & Struchiner, C. J. (1996) Am. J. Epidemiol. 144, 83-97.
17. O’Hagan, J. J., Hernán, M. A., Walensky, R. P. & Lipsitch, M. (2012) AIDS 26, 123.
18. Gomes, M. G. M., et al. (2014) PLOS Pathog. 10, e1003849.
19. Gomes, M. G. M., Gordon, S. B. & Lalloo, D. G. (2016) Vaccine 34, 3007.
20. Langwig, K. E., et al. (2017) mBio 8, e00796-17.
21. Pessoa, D., et al. (2016) PLOS Comput. Biol. 10, e1003773.
22. King, J. G., Souto-Maior, C., Sartori, L. M., Maciel-de-Freitas, R. & Gomes, M. G. M. (2018) Nat. Commun. 9, 1-8.
23. Gomes, M. G. M., et al. (2020) medRvix 10.1101/2020.04.27.20081893.

Wednesday 13 October 2021 - Bruce Edmonds (Manchester Met)
Staging abstraction from a complex model of voter turnout
Simpler models (e.g. mathematical models that are analytically tractable) can rigorously determine the overall behaviour of a model, but the approximations and assumptions necessary to make these solvable can make the connection to what is being modelled weak. Complicated simulation models can more directly represent observed processes (as they are understood), i.e. are relevant, but are difficult to rigorously understand. Thus, it is increasingly common to use a combination of simulation and analytical models when trying to understand complex systems. This talk discusses some work which extends and formalises this approach, staging the abstraction into a sequence of models, starting with a complicated, descriptive model (that represents) but then progressively simplifying this into 3 more stages with an analytical model at the 'top'. This attempts to obtain both rigour and relevance of the modelling but at the cost of a more complicated 'pile' of models and a lot more work. The particular target that this work focussed on was "why people bother to go out and vote", which was part of a collaborative grant between the department of Theoretical Physics and the Cathie Marsh Centre (for quantitative social science) at the University of Manchester, with us modellers, at the Centre for Policy Modelling, Manchester Met. University being the bridge between the two.