News

Torsten successfully defended his PhD on Thursday, November 12th with summa cum laude! In his PhD, Torsten developed two important methods for reverse engineering regulatory networks. One method, called response logic, allows to reverse engineer the topology of a network from perturbation data (Gross et al, 2019). A second method allows to identify which perturbation experiments would be optimal to quantiatively describe the network (Gross et al. 2020). The works were presented at the ISMB 2019 and 2020, and received prices for best student paper and best talk, respectively! Torsten will continue his career in London to work on machine learning application in health and biotechnology! We wish Torsten all the best for his future career!

we would like to announce our next "LAP-Life after PhD" seminar which will take place on November 17 at 4:30 p.m. as an online seminar via zoom.

Our speaker will be Dr. Anncharlott Berglar, who has done her PhD at Institut Pasteur followed by an MA in Scientific Illustration and is now a freelance scientific illustrator at SciVisLab.

Date: Nov, 17th

Time: 4:30 p.m.
Venue: Zoom

For more information please visit the IRTG2403 website. You will find regular updates at https://www.regulatory-genome.hu-berlin.de/en/events/lectures/lap-series.

Colorectal Cancer (CRC) is the 3rd most commonly occurring cancer world-wide. The past two decades of intense research have indeed advanced our understanding of the genetics underlying the formation of an adenoma (benign tissue) and carcinoma (cancerous tissue) of CRC, albeit utilizing mainly unmatched patient cohorts of adenoma and carcinoma. However, although key driver DNA variants that distinguishes both entities have been well established in the field, the determinant and earliest variant that selects an adenoma to progress to a carcinoma remains unknown. Mamlouk et. al. investigated this with a unique cohort of matched patient samples consisting of polyps carrying adenomas captured at the transition stage to carcinoma. We identified that key alterations in TP53 and chromosome 20 gain are early events driving the progression towards carcinoma. They were not only found shared between adenoma-carcinoma pairs, but also, distinguished low-grade from more high-grade adenoma. This highlights the major finding of the publication that the molecular progression, that is DNA alterations such as mutations and copy number changes, are uncoupled from the histological progression within these tumors. We further expanded on the heterogeneity present within the polyps by performing clonal deconvolution analysis using mutational data from multi-regional tissue isolation. We showed that selective pressure occurs at both adenoma and carcinoma tissue and subclonal populations are further evident within adenoma tissue long after its progression to a carcinoma.

Mamlouk, S., Simon, T. et al., Malignant transformation and genetic alterations are uncoupled in early colorectal cancer progression

https://bmcbiol.biomedcentral.com/articles/10.1186/s12915-020-00844-x

In the field of computer vision, probabilistic convolutional neural networks, which predict distributions of predictions instead of point estimates led to many recent advances. Besides state of the art benchmark results, these networks made it possible to quantify local uncertainties in the predictions. These were used in active learning frameworks to target the labeling efforts of specialist annotators or to assess the quality of a prediction in a safety-critical environment. However, for instance segmentation problems, which aim at separating different objects (e.g. cells) from one another, these methods are not frequently used so far. We seek to close this gap by proposing a generic method to obtain model-inherent uncertainty estimates within proposal-free instance segmentation models. Furthermore, the quality of the uncertainty estimates is analyzed with a metric adapted from semantic segmentation, which seeks to separate objects based on their class (e.g. epithelial cells or other cells). The method is evaluated on a dataset that contains C.elegans brightfield microscopy images, where it yields competitive performance while also predicting uncertainty estimates that carry information about object-level inaccuracies like false splits and false merges. These uncertainty estimates are then used in a simulation study to guide proofreading efforts.

The manuscript was accepted for the Bio Image Computing Workshop within ECCV2020 and is available at arXiv

Rumberger, J.L., Mais, L., Kainmüller, D. Probabilistic Deep Learning for Instance Segmentation

https://arxiv.org/abs/2008.10678

Epithelial ovarian cancer (EOC) is the leading cause of death within gynecological cancers in the developed countries. Due to the lack of specific symptoms, EOC is often detected at an advanced stage with a five-year survival rate less than 40%. However, 25% of EOC patients are diagnosed in early stage (I-II), where the disease is often cured by surgery alone, or in combination with platinum-based chemotherapy. Even though the prognosis of patients with FIGO stage I-II increases dramatically with treatment, with five-year survival rates between 80–90%, some subgroups of early-stage EOC will relapse and 20–30% of these patients will finally succumb to the disease. Nevertheless, the optimal clinical management is still a controversial debate and patients with early-stage high-grade serous EOC might be over-treated which could potentially result in complications after radical surgical management and an increase in toxicity of chemotherapy. Hence, it is of utmost importance to identify novel diagnostic markers for this patient cohort in order to improve the risk assessment of tumor recurrence. Here, we have applied MALDI-imaging mass spectrometry (MALDI-IMS), a new method to identify distinct mass profiles including protein signatures on paraffin embedded tissue sections. In search of prognostic biomarker candidates, we compared proteomic profiles of primary tumor section from early-stage HGSOC patients with either recurrent or non-recurrent disease, and were able to identify a discriminative peptide signature to predict clinical outcome and treatment extent for patients with early-stage HGSOC.

Kulbe H., ... , Kassuhn W., et al., Discovery of Prognostic Markers for Early-Stage High-Grade Serous Ovarian Cancer by Maldi-Imaging, Cancers 2020

https://www.mdpi.com/2072-6694/12/8/2000/htm