Dienstag, 19.10.2021, 12-13 Uhr - in H8
Jonas Bauer, M.Sc.
Universität Bielefeld
Uncertainty quantification in scRNA-Seq
Scientists today are still puzzled by many diseases that are related to changes on the level of gene expressions, such as cancer. Studying, for instance, blood or tissue samples can help to understand the underlying microbiological mechanism, the progression of a disease itself and finally to deduce appropriate treatments. Therefore it is necessary to quantify gene expressions in a sample, and distinguish healthy from infected cells adequately. However, one cannot observe and count these molecules directly. Instead, several intermediary steps are necessary which lead to millions of duplicated and much shorter RNA sequences. This talk discusses the impact of the single cell RNA-Seq set-up on the estimation accuracy and the various sources of uncertainty that are involved. Four mapping approaches are compared and a Bayesian framework for the data generating process as well as for Hamiltonian-Monte-Carlo-based sampling is presented.
The talk will be held in English.
Dienstag, 02.11.2021, 12-13 Uhr - Vortrag über zoom
Prof. Dr. Annika Hoyer
Universität Bielefeld
Methods for the meta-analysis of ROC curves – A statistical challenge
While methods for meta-analysis, i.e. the weighted summary of the results of several studies, are meanwhile widespread in the context of intervention studies, there is still a need for innovative statistical approaches for meta-analysis of diagnostic accuracy studies. This can be attributed to the complex underlying data with an at least bivariate outcome consisting of sensitivity and specificity. Sensitivity and specificity refer to the conditional probabilities of a positive or negative test result in the population of diseased and non-diseased, respectively. A particular challenge arises when individual studies report not only a pair of sensitivity and specificity, but complete receiver operating characteristic (ROC) curves in which sensitivity and specificity are evaluated at several different diagnostic thresholds. One way to appropriately deal with these data is to adapt statistical methods from survival analysis, such as the application of bivariate time-to-event models for interval-censored data. In the talk, newly developed, innovative methods for the complex situation of meta-analysis of diagnostic studies, especially in case of ROC curves, will be presented and discussed.
The talk will be held in English.
Dienstag, 16.11.2021, 12-13 Uhr - in H8
Julia Dyck, M.Sc.
Universität Bielefeld
Parameter uncertainty estimation for exponential semi-variogram models
Parametric semi-variogram modelling is a frequently used application in spatial data analysis. One commonly chosen parametric form is the exponential semi-variogram model. While proper inference should contain uncertainty quantification of parameter estimates, the estimation of parameter standard errors for parametric semi-variogram models comes along with some challenges, like the natural correlation structure within the data and the two-step process required to fit a semi-variogram model. The focus of this talk is on parameter standard error estimation for exponential semi-variogram models in the context of datasets with around 500 to 2000 observations and little control over the sampling design. These framework conditions are motivated by spatial data analyses in epidemiology. After an introduction to the concept of the semi-variogram, previously developed methods as well as a new modification are presented. Their performance within simulated data scenarios is summarized and discussed.
The talk will be held in English.
Dienstag, 30.11.2021, 12-13 Uhr - Vortrag über zoom
Dr. Annette Möller
Universität Bielefeld
Vine copula based post-processing of ensemble forecasts for temperature
Today weather forecasting is conducted using numerical weather prediction (NWP) models, consisting of a set of differential equations describing the dynamics of the atmosphere. The output of such NWP models are single deterministic forecasts of future atmospheric states. To assess uncertainty in NWP forecasts so-called forecast ensembles are utilized. They are generated by employing a NWP model multiple times for distinct variants. However, as forecast ensembles are not able to capture the full amount of uncertainty in an NWP model, they often exhibit biases and dispersion errors. Therefore it has become common practise to employ statistical post-processing models which correct for biases and improve calibration. We propose a novel post-processing approach based on D-vine copulas, representing the predictive distribution by its quantiles. These models allow for much more general dependence structures than the state-of-the-art Ensemble Model Output Statistics (EMOS) model and is highly data adapted. Our D-vine quantile regression approach shows comparative predictive performance in studies of temperature forecasts over Europe with different forecast horizons based on the 52-member ensemble of the European Centre for Medium-Range Weather Forecasting (ECMWF). Specifically for larger forecast horizons the method improves over the benchmark EMOS model.
The talk will be held in English.
Dienstag, 14.12.2021, 12-13 Uhr - Vortrag über zoom
Jonas Bauer, M.Sc., Qingchuan Sun, M.Sc., Sophie Thiesbrummel, M.Sc., Julian Wäsche, M.Sc. and Houda Yaqine, M.Sc.
Universität Bielefeld
Everything but a DREAM: Prediction of COVID-19 progression in American children
For almost two years, the COVID-19 pandemic has been posing a serious challenge to almost every area of our lives, but especially in the medical field. While older patients are in general at risk of a more severe progression, infected children are mostly asymptomatic or experience only mild symptoms. Nevertheless, some pediatric patients progress to severe COVID-19 infections with life-threatening courses of the disease. Hence, a collaboration of different US health authorities has launched the “Pediatric COVID-19 Data Challenge” in order to develop computational models for the prediction of severe infections in children. The tasks consist of risk prediction of hospitalization for children who have been tested positive for COVID-19 and the subsequent assessment of the risk for intensive care, given the patient has been hospitalized. In this presentation, we provide a profound overview of the OMOP data format as well as the N3C enclave, discuss obstacles that we have been facing, and present approaches to tackle these tasks.
The talk will be held in English.
Dienstag, 11.01.2022, 12-13 Uhr - Vortrag über zoom
Dr. Christian Rudloff
AIT Austrian Institute of Technology GmbH
Mobility surveys beyond stated preference: introducing MyTrips, an SP-off-RP survey tool and the estimation of Latent Class Mode Choice Models
When introducing new mobility offers or measures to influence traffic, stated preference (SP) surveys are often used to assess their impact. In SP surveys, respondents do not answer questions about their actual behaviour, but about hypothetical settings. Therefore, answers are often biased. To minimise this hypothetical bias, so-called stated preference-off-revealed preference (SP-off-RP) surveys were developed. They base SP questions on respondents’ revealed behaviour and place unknown scenarios in a familiar context. Until now, this method was applied mostly to scenarios investigating the willingness to pay. In MyTrips, SP-off-RP is applied to more complex mode or route choice problems, which require the calculation of routes. The MyTrips survey tool was developed to collect SP-off-RP data based on respondents’ actual mobility behaviour. SP questions are based on alternatives to typical routes of respondents, which are calculated on the fly with an intermodal router. MyTrips includes a larger survey and collects mobility diaries for one day representing respondents’ daily routine, calculates alternative routes and creates SP questions based on a Bayesian optimal design. In addition, the SP-off-RP data is used to estimate Latent Class models to investigate to model mode-choice behaviour of the respondents. Results of two case studies show a difference in response behaviour between SP and RP settings and suggest a reduction of hypothetical bias.
The talk will be held in English.
Dienstag, 25.01.2022, 12-13 Uhr - Vortrag über zoom
Aktuelle Forschungsthemen im ZeSt