Institute of Medical Biometry and Statistics (IMBI)

Bioconductor R-package for Data Normalization

This package provides a normalization method for omics data where a few features are highly expressed. Then the traditional quantile normalization that is based on ranking proteins is not applicable any more. We developed a mean balanced quantile normalization that is also applicable in such settings.

Installation: https://www.bioconductor.org/packages/release/bioc/html/MBQN.html

Publication: Brombacher, E., Schad, A., & Kreutz, C. (2020). Tail‐Robust Quantile Normalization. Proteomics, 20(24), 2000068.

Analysis of Single Cell Data

In this project, we compare the performance of statistical models and tests as well as machine learning methods (e.g. clustering) for processing of single cell data, e.g. from scRNA-seq or mass cytometry. We also establish pipelines which are robust against the choice of configuration parameters.

Responsible: Eva Kohnert, Ariane Schad, Clemens Kreutz

Our collaborators: Stefan Reinker (Novartis), Florens Lohrmann, Philipp Henneke

Modelling Methods in Systems Biology

In this project, we develop and extend existing methodology for establishing mathematical models of biochemical processes in living cells. We focus on

• development of reliable and robust optimization methods for parameter estimation
• deriving a valid statistical methodology for parameter and prediction uncertainties as well as for identifiablity and observability analyses
• establishing strategies like L1 regularization for deriving small minimal models
• approximating differential equation models by functions as a prerequisite for multi-scale modelling
• evaluate deep learning approaches in the context of ODE modelling

Responsible: Lukas Refisch, Rafael Arutjunjan, Clemens Kreutz

Our collaborator: Jens Timmer, Andreas Raue (Merrimack Pharm.)

Benchmarking in Systems Biology

In this project, we evalute the performance of existing computational methods for mathematical modelling.

• We compare the performance of optimization methods which are applied for model calibration/parameter estimation
• We establish benchmark problems which can be used to assess modelling methods
• We develop and extend methodology for performing reliable benchmark studies
• We establish approaches for generation of simulation data realistically

Link: Benchmark models on github

Responsible: Lukas Refisch, Janine Egert, Clemens Kreutz

Our collaborators: Jens Timmer, Jan Hausenauer

Benchmarking Wiki

In this project, we started to develop a wiki as a repository for published benchmark studies. We intend to collect insights from literature obtained by comparing the performance of computational approaches.

The benchmarking wiki also offers the opportunity to publish (possibly small or unpublished) own results.

Everybody is welcome to contribute!

Link: Benchmark Wiki

Responsible: Clemens Kreutz