CMMC: Bozek, Katarzyna - assoc. RG 33

Introduction

We are interested in resolving large data challenges in biology and medicine. With the use of deep learning we aim at developing new data-driven approaches to study of image and video data in biology. Our theoretical interests lie in using machine learning to find appropriate data representations for resolving specific scientific questions. We apply and develop computational and analytical solutions to questions in cancer and ageing research.

Located in the interdisciplinary environment of the CMMC, our group is involved in multiple close collaborations with partners from experimental and medical backgrounds. Strongly applied in biomedical research, we search for ways in which computational and in particular machine learning methods can advance resolving specific medical and biological questions. Currently we pursue two major projects.

Systems medicine of triple negative breast cancer

Highly differentiated diseases require large-volume, high-resolution datasets and novel analytical approaches to find recurring patterns in large spectrum of genetic and molecular variation. In collaboration with experimental groups at the Charité University Hospital Berlin we address the question of high molecular heterogeneity in Triple Negative Breast Cancer (TNBC). This most aggressive breast cancer subtype with high rates of recurrence and mortality offers currently no therapy options.

In this project we expand over the existing knowledge of the genetic heterogeneity between and within TNBC tumors, by investigating the differences in their transcriptome, proteome and morphology as well as the extracellular matrix components and immune cell environment. Our analysis is based on patient samples from prospective clinical trials at multiple levels. We strive to establish a novel type of biomarkers that instead of individual molecule levels will integrate multiple types of information and will be quantified with the use of machine learning and mathematical models.

Figure 1

Fig. 1. We combine different types of large-scale biological information such as sequence data, protein levels and also importantly visual information. Using machine learning we build quantitative representations of image data which allows to perform integrative analysis of the highly diverse biological information combining morphological and dynamic aspects of systems with the underlying molecular states of tissues.

C. elegans phenotyping

C. elegans, a tiny nematode worm, is used to study a broad range of questions in biology, from diseases to neural function. This apparently simple organism shows a broad repertoire of behaviors incomprehensible to human observer. These behaviors might be representative of its health and disease phenotype.

We employ deep learning methods to quantify and search for distinct motion patterns representative of worm molecular phenotype. The grand challenge of this project involves finding ways to represent worm posture and dynamics. Inspired by methods for language processing we aim to find analogous meaningful representations of words and sentences in the language of worm behavior.

Figure 2

Fig. 2. C. elegans posture is represented via a low-dimensional numeric vector obtained through an autoencoder architecture (A). Representations calculated for a series of postures captured in a video conveniently capture worm motion dynamics (B). The RNN architecture used for this purpose was originally designed for language translation (Sutskever et al. 2014). As translation of sentences word-by-word might produce incorrect results, language processing architectures derive an internal representation on the basis of which sentence in the target language is reconstructed. Internal representations of video sequences obtained this way allow to group together similar behaviors as well as to quantitatively assess behavioral differences of genetically different C. elegans strains (C).

Perspectives

The recent progress of both imaging and image analysis techniques are currently opening ways to elevate image data to a new scale in biological research. Whether in diagnostics of cancer or phenotyping of model organisms, we aim to establish methods and approaches allowing for broad, data scientific use of visual data in biology.

Lab Website

For further information, please check the Bozek Laboratory for Data Sience of Images' webpage.

Publication Record on PubMed - Katarzyna Bozek

Publications generated during 1/2023-12/2025 with CMMC affiliation

2024 (up to June)

Naji H, Sancere L, Simon A, Buttner R, Eich ML, Lohneis P, and Bozek K (2024). HoLy-Net: Segmentation of histological images of diffuse large B-cell lymphoma. Comput Biol Med170, 107978. doi:10.1016/j.compbiomed.2024.107978.
Lukomski L, Pisula J, Wagner T, Sabov A, Grosse Hokamp N, Bozek K, Popp F, Kann M, Kurschat C, Becker JU, Bruns C, Thomas M, and Stippel D (2024). First experiences with machine learning predictions of accelerated declining eGFR slope of living kidney donors 3 years after donation. J Nephrol. doi:10.1007/s40620-024-01967-y.

2023

Butt L, Unnersjo-Jess D, Reilly D, Hahnfeldt R, Rinschen MM, Bozek K, Schermer B, Benzing T, and Hohne M (2023). In vivo characterization of a podocyte-expressed short podocin isoform. BMC Nephrol 24, 378. doi:10.1186/s12882-023-03420-x.
Deserno M, and Bozek K (2023). WormSwin: Instance segmentation of C. elegans using vision transformer. Sci Rep 13, 11021. doi:10.1038/s41598-023-38213-7.
Jung JO, Pisula JI, Bozek K, Popp F, Fuchs HF, Schroder W, Bruns CJ, and Schmidt T (2023). Prediction of postoperative complications after oesophagectomy using machine-learning methods. Br J Surg 110, 1361-1366. doi:10.1093/bjs/znad181.
Pisula JI, Datta RR, Valdez LB, Avemarg JR, Jung JO, Plum P, Loser H, Lohneis P, Meuschke M, Dos Santos DP, Gebauer F, Quaas A, Walch A, Bruns CJ, Lawonn K, Popp FC, and Bozek K (2023). Predicting the HER2 status in oesophageal cancer from tissue microarrays using convolutional neural networks. Br J Cancer 128, 1369-1376. doi:10.1038/s41416-023-02143-y.
Unnersjo-Jess D, Butt L, Hohne M, Sergei G, Fatehi A, Witasp A, Wernerson A, Patrakka J, Hoyer PF, Blom H, Schermer B, Bozek K, and Benzing T (2023). Deep learning-based segmentation and quantification of podocyte foot process morphology suggests differential patterns of foot process effacement across kidney pathologies. Kidney Int 103, 1120-1130. doi:10.1016/j.kint.2023.03.013.
Vom Stein AF, Rebollido-Rios R, Lukas A, Koch M, von Lom A, Reinartz S, Bachurski D, Rose F, Bozek K, Abdallah AT, Kohlhas V, Saggau J, Zolzer R, Zhao Y, Bruns C, Brockelmann PJ, Lohneis P, Buttner R, Haupl B, Oellerich T, Nguyen PH, and Hallek M (2023). LYN kinase programs stromal fibroblasts to facilitate leukemic survival via regulation of c-JUN and THBS1. Nat Commun 14, 1330. doi:10.1038/s41467-023-36824-2.

Bozek, Katarzyna - assoc. RG 33

Data Science of Bioimages

Prof. Dr. Katarzyna Bozek