Current Research Projects
isba |
Sponsors
Actuaries and STatisticians endeavour to design innovative, inclusive insurance products in a changing RISK landscape (ASTeRISK)
Actuaries design and value risk transfers by analyzing insurance data sets with a sophisticated statistical toolbox. Current practice faces technical challenges in the design and estimation of risk models from fine-grained data. At the same time, trust in modern-day insurance is under pressure. Policyholders and regulators expect value creation from the collected granular data in the form of better insurance products and a wider coverage of (new types of) risks. New forms of mutual insurance (e.g., peer-to-peer covers) recently emerged outside traditional insurance, inspired by evolutions in the sharing economy. In response to these challenges our project aims to shape the statistical toolbox necessary for building valid, reproducible and viable risk models for the granular data collected in a variety of key actuarial tasks. Building on solid methodological foundations our project innovates by explicitly incorporating a dynamic prediction and prevention view into these risk models. Moreover, we aim to contribute to a better, more inclusive and fair access to insurance with the design of a wide array of new risk-sharing solutions and the development of novel quantitative tools to spot (proxy) discrimination in insurance pricing. Finally, the project targets tailored statistical models to assess actual riskiness and to refine the implementation of the recent ‘right to be forgotten’ that should improve the access of cancer survivors to essential insurance products.
Promoters: Katrien Antonio (KU Leuven) and Donatien Hainaut (UCLouvain)
UCLouvain Team: Academics: Michel Denuit, Anouar EL Ghouch, Donatien Hainaut, Catherine Legrand.
KU Leuven Team: Academics: Katrien Antonio, Gerda Claeskens, Jan Dhaene, Ingrid Van Keilegom.
Researchers: Aurèle Bartolomeo, Charlotte Jamotton, Patricia Ortega-Jimenez
Website : https://eos-asterisk.netlify.app/
Learning structures and patterns from multivariate extremes
Learning new knowledge from data at the outer regions of a sample is challenging because, by definition, the amount of such data is scarce and model assumptions have a disproportionately large impact on the conclusions. Our overall objective is to leverage and develop techniques from multivariate extreme value analysis to enrich the tool-set for analyzing and learning from the extremes of a set of data. Specifically, we aim to extend current models for multivariate peaks-over-thresholds to the case that such peaks occur in small clusters of variables and we aim to develop methods to learn the graphical structure of dependence relations between the extremes of the components of a high-dimensional random vector. The methods developed will be put into practice in the analysis of data involving weather and climate extremes.
Promoters : Anna Kiriliouk (UNamur) and Johan Segers
Researcher: Anas Mourahib
ETHIAS Chair 2023-2026
Fully funded Pension Systems
In the current context of an ageing population, with falling birth rate and increasing life expectancy, the adaptation of pension schemes is a major societal challenge for the coming decades. Trying to combine financial sustainability and social adequacy of pensions, many European countries have undertaken changes to their pension systems. Belgium is not far behind, with the public authorities having adopted various measures The government’s objective is to substantially continue this series of reforms in the coming years. The Ethias Chair focuses on supplementary pensions and pursues two objectives: - to consider the design of fair and sustainable pension systems; - to contribute to maintaining a transdisciplinary platform for research on pensions at UCLouvain (law, actuarial sciences, Economics).
Promoter: Pierre Devolder
Imperfect Data : From Mathematical Foundations to Applications in Life Sciences (IMAL)
We are witnessing a period of time where the data collection potential has increased exponentially. The cautionary tale of this big data era is that large amounts of data do not necessarily contribute to an increment in our knowledge about the underlying phenomenon. One of the principal reasons for this is that even though one would desire to measure a characteristic for a subject, in many instances one can only get an approximate measurement due to difficulty in obtaining the direct measurement of the desired phenomenon (e.g. tumor size), non-replicability across instances (e.g. blood pressure), necessity to obtain numerous measurements rapidly, sometimes at the cost of accuracy. As a result, many modern observed markers are proxies for the real data because invasive, costly or too complex methods would be required to obtain accurate measurements. In this project we study how one can correct for different types of imperfect data when building statistical models with a focus on applications coming from life sciences. Imperfect data appear in different contexts, structures and models, and this project focuses on two common settings which regularly suffer from imperfect data: data in a regression context with imperfectly measured explanatory variables (Theme 1) and highdimensional or functional data with measurement error (Theme 2).
Promoters : Catherine Legrand (porte-parole, UCLouvain), Anouar El Ghouch (UCLouvain), Philippe Lambert (UCLouvain / ULiège), Eugen Pircalabelu (UCLouvain), Germain Van Bever (UNamur), Ingrid Van Keilegom (UCLouvain / KU Leuven).
ARC project
PDR projet de recherche- FNRS / FRESH / FRIA 2023-2025
The empirical angular Wasserstein distance: learning from multivariate extremes
The angular measure of a probability distribution on Euclidean space describes the directions in space in which points far from the centre of the distribution (far in an asymptotic sense) are most likely to be found. The empirical angular measure is a non-parametric estimator that underlies a variety of methods in extreme value analysis, which is the branch of statistics concerned with atypical sample values, such as sample maxima, high-thresholds excesses, and multivariate versions thereof. The aim of the project is to contribute to the sampling theory of the empirical angular measure by investigating the finite-sample and asymptotic distributions of the Wasserstein distance between the empirical and true angular measures, or empirical angular Wasserstein distance in short. For bivariate probability distributions, the angular measure is supported on the one-dimensional unit sphere, and the asymptotic distribution of the empirical angular Wasserstein distance will be found through empirical process theory. In higher dimensions, developing such asymptotic distribution theory will also be possible for angular measures with finite support thanks to the Kantorovich duality. For general angular measures however, the focus will rather be on rates of convergence and concentration bounds for the empirical angular Wasserstein distance, in line with the literature on the Wasserstein distance between the usual empirical and true probability distributions. The results will be based on concentration inequalities for the empirical angular measure, derived in turn from such inequalities for rare events. Throughout, various statistical methods for multivariate extremes based on the angular measure will be considered.
Promotor: Johan Segers, Researcher: Stéphane Lhaut
Cluepoints 2016-2026
Central Statistical Monitoring of Clinical Trials
In 2012-2014, a researcher was hired full-time at UCLouvain to collaborate with other researchers both from private sector and other universities in a project financed by BioWin on the development and validation of a platform for statistical monitoring of data quality of clinical trials.
This collaboration has continued on an informal basis, in particular to follow-up on publications. In the mean time a spin-off company (CluePoints) was created and its R&D department has steadily grown.
Since 2016, part-time (20%) research at UCLouvain has been funded by CluePoints. Activities of the researcher are varied and include participation in research meetings, preparation of publications as well as development of new methods. Validation of methods is conducted both on simulated data and on case-studies.
Promotor: Catherine Legrand, Researcher: Lieven Desmet
Fondation Francqui: Start-Up Grant
Estimating Direct and Indirect Effects of the Menstrual Cycle on the Glycemic Response from in-situ Data.
The prevalence of (pre-)diabetes is continuously increasing. An important health indicator is the magnitude of the glycemic response after a meal. Associations between reproductive conditions (e.g., PCOS) and insulin resistance suggest that reproductive hormonal levels might alter this response. This project aims at quantifying the direct and indirect (e.g., through changes in diet or physical activity) effects of the menstrual cycle on the glycemic response. It relies on high-resolution longitudinal data from 80+ menstruating healthy women that were collected in situ (i.e., in normal living conditions) through digital health devices and apps for four weeks. Bayesian models of the glycemic response to various inputs (food, sport, sleep) will be fitted to this data and evaluated. These models will need to account for uncertainties inherent to the self-reported nature of the input data regarding their timing, amounts, and duration.
Promotor: Laura Symul
VIBRANT (Vaginal lIve Biotherapeutic RANdomized Trial)
Globally, approximately 30% of women have bacterial vaginosis (BV). Antibiotic treatment is frequently followed by a lack of colonization with beneficial lactobacilli and recurrence of BV frequently follows antibiotic treatment. Recently, a single-strain Lactobacillus crispatus live biotherapeutic product (LBP) has shown promising results in phase I and phase II trials but highlighted that only about a 3rd of participants receiving the LBP successfully colonized with L. crispatus. One hypothesis for this low proportion is that a single-strain LBP lacks functional diversity to adapt to any host-specific environment. Consequently, this project proposes to test a multi-strain Lactobacillus crispatus LBP in the VIBRANT trial. VIBRANT is a randomized, placebo-controlled trial of multi-strain Lactobacillus crispatus vaginal live biotherapeutic product in women with BV in the United States (US) and South Africa (ZA). The live biotherapeutic products (LBP) are vaginal tablets: LC106, containing 6 L. crispatus strains and LC115, containing 15 L. crispatus strains, both with 2 x 109 colony forming units
(CFU) per dose.
Promoter: Laura Symul
Identifying drivers of genital inflammation and HIV acquisition in women living in sub-Saharan Africa.
The female genital tract (FGT) plays a pivotal role in the acquisition of HIV infection in women and heterosexual transmission to men. Although most new infections occur in women following exposure to HIV through the mucosal membranes of the FGT, there remain significant gaps in our understanding of the local mucosal factors influencing susceptibility. Our previous work demonstrated that, relative to vaginal communities dominated by Lactobacillus crispatus, high diversity vaginal bacterial communities comprised of mixed anaerobes with low Lactobacillus abundance are associated with elevated levels of inflammatory chemokines and cytokines in genital secretions, increased numbers of activated CD4+ T cells (“HIV target cells”) in the cervix, and a greater than four-fold increased risk of HIV acquisition. Importantly, these high risk vaginal bacterial communities are present in two-thirds of young, healthy, black South African women in our FRESH
(Females Rising through Education, Support and Health) cohort, suggesting that this risk factor is highly prevalent among South African women of reproductive age. Furthermore, vaginal microbiota and genital inflammation have been shown to significantly reduce the efficacy of vaginal microbicides for preventing HIV.
These findings underscore the critical need to understand the mechanisms by which vaginal microbiota increase genital inflammation and HIV acquisition risk, particularly for women living in sub-Saharan Africa. To address these knowledge gaps, we will utilize 3600 samples from 1200 FRESH participants with longitudinal blood and genital sampling, including pre- and post-infection samples from 102 women who went on to acquire HIV. In Aim 1.1 we will generate multi-omics datasets characterizing vaginal microbiota utilizing metagenomics, metatranscriptomics, fungal internally transcribed spacer (ITS) sequencing, culturomics, microbial whole genome sequencing (WGS), and metabolomics. We will complement this in Aim 1.2 with a comprehensive characterization of the host FGT by assessing genital inflammatory chemokines and cytokines, cervical flow cytometric cell phenotype, and cervical single-cell RNA sequencing (scRNAseq) profiles, combined with in-depth behavioral and demographic metadata. These complex multi-omics datasets will then be integrated in Aim 1.3 following critical quality control, transformation, and imputation of data from each modality. Modern supervised integration techniques will be used to identify multi-omics profiles predictive of genital inflammation and HIV acquisition. Candidate vaginal bacterial strains, functions, and/or metabolites will then be validated in Aim 2 using innovative in vitro and in vivo models, including an organ-on-a-chip model co-developed by the co-PI, cervical and vaginal organoids, and a humanized mouse model. Our preliminary data indicate promising candidates and hypotheses that can already be tested directly. Overall, this project will identify specific bacterial strains and molecules, along with host cellular lineages and sensing pathways, that mediate microbiota-induced genital inflammation and increased HIV acquisition risk. It will also validate much-needed novel model systems that test the interactions of vaginal microbiota, host cells, and HIV to provide critical resources to develop the mechanistic understanding needed to advance the field.
Promoter: Laura Symul