Said Alkildani
Spatial Composition of the Tumor Microenvironment as Predictive Biomarker for Immunotherapy Response in Pl-L1-high Non-small-cell Lung Cancer
Supervisors: Philip Bischoff, Nils Blüthgen
Non-small cell lung cancer (NSCLC) is the most common cause of cancer-related deaths worldwide. Diagnosis is most often made in advanced stages where surgical intervention is deemed unproductive and systemic therapy is needed. A common immune evasive mechanism of tumor cells is the hijacking of the regulatory PD-1/PD-L1 interaction that inhibits T cell activity. Immune checkpoint inhibitors (ICIs) targeting this interaction are a common first-line intervention. Even though sensitivity to ICIs is determined histologically prior to treatment by PD-L1 immunohistochemistry, a substantial number of PD-L1-high patients do not respond to therapy as expected or suffer from recurrence within two years of treatment. My project aims to understand the mechanisms behind this phenomenon and to develop a more precise clinical routine that comprehensively captures the state of the tumor microenvironment. I am using spatial single-cell gene expression profiling of FFPE-fixed NSCLC samples to uncover novel aspects of tumor evolution from early to advanced PD-L1-high milieu, as well as to understand the underlying mechanisms behind primary and acquired resistances. Additionally, I am coupling my research with immunofluorescence imaging, genomic profiling, publicly available data integration, and single-cell RNA sequencing.
Siddharth Annaldasula
Deconvoluting the Tumor Microenvironment using Somatic Mutations
Supervisors: Kirsten Kübler, Dieter Beule
Somatic mutagenesis is a natural process that occurs in all tissues, where mutations accumulate at varying rates depending on the tissue type. In cancer, mutations in “driver genes” transform healthy cells (“cell-of-origin”) into a precancerous and eventually a cancerous state. Cancer cells are embedded in the tumor microenvironment (TME), which also consist of healthy tissue cells, immune cells, and stromal cells. Tumor samples are often obtained using bulk sequencing and are assigned a tumor purity value, which does not provide a qualitative characterization of the TME. As the aggregated somatic mutational burden within a bulk sample reflects the underlying mixture of cells, I aim to leverage somatic mutational counts to computationally deconvolute the cellular composition.
Rosario Astaburuaga
Understanding the signalling pathways governing cell fate decisions after radiation-induced DNA damage in HN cancer cells
supervisors: Nils Blüthgen, Kirsten Lauber, Hanspeter Herzel
Ionizing radiation is the main treatment strategy for head and neck (HN) cancers, as it can damage the DNA of cancer cells. As HN tumours are usually highly heterogeneous, after DNA damage some cells easily die (i.e, radio-sensitive cells), but others resist to die and survive (i.e., radio-resistant cells). The mechanisms governing the fate of cells after radiation-induced DNA damage are not well understood. This is a major clinical problem, as determines the success of radiation therapy and therefore the likelihood of developing a tumor relapse. We established a model system of intra-tumoral heterogeneity and divergent cell fate decisions by studying genetic subclones with different responses to radiation. To understand their differential behaviour in terms of signalling dynamics, we performed time-course mass cytometry (CyTOF) analyses of irradiated and non-irradiated cells. We so far uncovered plausible molecular mechanisms of resistance, that will be tested by perturbing signalling pathways.
I generally like data-driven analyses, which allows me to come up with new hypothesis that can be further tested. I’m interested in the DNA-damage response, specifically after ionising radiation, and the involvement of p53 and MAPK pathways. I enjoy single cell analyses, understanding the data by disentangling different sources of covariance like cell cycle, cell volume, and other cell state dynamics.
Finnja Becker
Development and identification of breakpoint-based mutational signatures
Supervisors: Anton Henssen, Hanspeter Herzel
My research focuses on mutational signatures, especially those derived from structural variants (SVs). SVs are large genomic rearrangements and are a major source of genomic instability in cancer. Current mutational signatures derived from SVs remain poorly characterised. They focus on variant type, length, and clustering, but overlook the DNA sequences surrounding breakpoint regions, despite evidence that these sequences can reveal mechanistic insights, such as RAG1/2 and PGBD5-mediated rearrangements. We propose a novel class of SV mutational signatures that incorporate local sequence context around breakpoints. Within the scope of my project, we are developing a computational pipeline to extract these breakpoint-based signatures from variant calling files, enabling the identification of known and novel sequence motifs linked to specific mutational processes. This approach complements existing SV signatures and enhances our understanding of the genomic mechanisms driving structural rearrangements in cancer.
Jonas Berger
Identification of novel immunotherapeutic targets in myeloid malignancies using single cell long-read sequencing
Supervisors: Livius Penter, Il-Kang Na, Silke Rickert-Sperling
My research focuses on identifying novel immunotherapeutic targets in myelodysplastic syndrome (MDS) and acute myeloid leukemia (AML). Recurrent mutations in the spliceosome provide opportunities for therapeutic exploitation of altered splicing profiles. By utilizing single cell long-read sequencing I aim to systematically identify aberrantly spliced gene isoforms in AML/MDS, including variants that give rise to immunogenic neoepitopes. By combining computational approaches with wet lab validation, I aim to identify neoantigens that provide novel targets for T-cell therapy in myeloid malignancies.
Roshni Biswas
Integrative clonal inference and lineage tracing using single-cell multi 'omics and machine learning
Supervisors: Jan-Philipp Junker, Leif Ludwig, Anton Henssen
Dileshi Bulathsinhala
Discovering molecular insights for combatting therapy resistance in neuroblastoma with computational modelling
Supervisors: Nils Blüthgen, Hanspeter Herzel
In the discipline of paediatric oncology, my project focuses on neuroblastoma. The overall long-term survival rate for patients with stage 4 neuroblastoma is poor, and this is primarily due to lethal relapses that result from therapy resistance and early metastasis. In order to combat therapy resistance, we require molecular insights to tumour evolution. I will apply computational modelling to analyse the impact of different RAS pathway mutations on the feedback and rewiring of the signalling pathway in order to understand resistance development. Our goal is to identify combinatorial treatments capable of sensitising highly resistant neuroblastoma cell lines.
Michael Eibl
Deciphering the combinatorial signaling in neural crest development across space and time
Supervisors: Stefanie Großwendt
Cell-cell communication (CCC) governs cellular behavior and fate in health and disease. In the developing embryo, multipotent neural crest cells respond to external cues to migrate along distinct paths and differentiate into a multitude of cell types, and thus provide an attractive model for studying the complex interplay of CCC, cell migration, and fate. Important signaling pathways have been characterized for some neural crest subpopulations. However, CCC is inherently multifactorial – even at the level of an individual cell –, and we are unaware of the combinatorial use of the full repertoire of signaling pathways by individual cells across the neural crest lineage. As local signaling occurs within cellular neighborhoods, CCC analysis requires considering spatial proximity between cells. While spatial transcriptomics retain the spatial context, substantially fewer transcripts are captured per cell than in single-cell RNA-sequencing (scRNA-seq). Recently, scRNA-seq was adapted to record gross spatial information. Here, we integrate this deep, spatially aware scRNA-seq with spatial transcriptomics data of the mouse embryo to identify combinatorial CCC of neural crest cells and their subpopulation-specific communication modules. Our approach will be applicable to other systems, including neural-crest-lineage derived cancers (e.g. neuroblastoma), and provide a blueprint for mapping comprehensive spatiotemporal communication patterns underlying cell fate in development and disease.
Mohamed Elmofty
Question Answering for Precision Oncology
Supervisors: Ulf Leser
Abeera Fatima
Coreness-Driven Node Embeddings: Advancing Functional Module and Biomarker Discovery
Supervisors: Martin Vingron, Ralf Herwig
My research in network biology focuses on developing advanced methods for decoding complex biological systems. Utilizing machine learning models akin to natural language processing techniques such as word2vec, I implement coreness-driven random walks to refine node embeddings. This approach enables the effective mapping of protein-protein interactions within a multi-dimensional framework. This methodological advancement significantly enhances our ability to identify functional modules, which are crucial for understanding cancer and other diseases. Ultimately, my research aims to improve biomarker discovery and therapeutic target identification, thereby advancing precision medicine and optimizing diagnostic and treatment strategies in oncology and broader healthcare domains.
Melanie Fattohi
Deciphering cell state transitions in embryo development and cancer
Supervisors: Stefanie Grosswendt, Laleh Haghverdi, Ulf Leser
My research focuses on investigating gene expression dynamics of cells in embryo development and the embryonal cancer neuroblastoma. To this end, I aim to develop computational single-cell methods for building gene regulatory networks to potentially gain new insights into normal and aberrant cell trajectories.
Annika Föhr
Identification of the cell-of-origin of cancers of unknown primary
Supervisor: Kirsten Kübler
I aim to identify the cell-of-origin of metastases with unknown primaries and to investigate whether they represent a distinct tumor entity, opening pathways for more effective treatment options.
Jannik Franzen
Understanding Colorectal Cancer Heterogeneity: Cell deconvolution from single-cell and spatial transcriptomics
Supervisor: Dagmar Kainmüller
Angelos Galanopoulos
Supervisors: Livius Penter, Nils Blüthgen
Samuele Garda
Neural Named Entity Normalization for the biomedical domain
Supervisors: Ulf Leser, Markus Schülke
My research focuses on the automatic extraction of structured information from the scientific literature. The aim is to accelerate the integration of new knowledge into resources crucial for biomedical research, including personalized cancer treatment. Specifically, I am interested in retrieving information regarding non-coding DNA sequences and the disease potential of their variants.
Mădălina Giurgiu
Extrachromosomal circular DNA structure heterogeneity in neuroblastoma
Supervisors: Anton Henssen, Knut Reinert, and Kerstin Haase
My research focuses on characterising extrachromosomal circular DNA structure diversity in neuroblastoma. To facility the study are developing methods for reconstructing these circular structures based on long-read nanopore data.
Aliki Grammatikaki
Dissecting neuroblastoma plasticity into developmental and cancer specific programs
Supervisors: Dieter Beule, Jan-Philipp Junker, Marvin Jens
Neuroblastoma (NB) is a pediatric cancer of the sympathetic nervous system, where transcriptional plasticity may drive relapse in high-risk patients. Using single-cell RNA sequencing (scRNA-seq), we study NB across patient data, healthy sympathoadrenal (SA) cells and ΝΒ cell-lines. Available data reveal pronounced tumor and cellular heterogeneity. Using non-negative matrix factorization (NMF), we identify robust transcriptional programs that capture developmental lineage stages, are shared between malignant and healthy cells, yet display differential activation.
Copy-number variation inference groups cells into distinct genetic subtypes. Cells of different subtypes display differences in activity and cell-to-cell variability of module activation. We aim to further delineate the genetic components of NB transcriptional heterogeneity from development plasticity, in order to either improve patient stratification and/or more clearly identify drivers of plasticity that could be targeted for treatment.
Fatemeh Habibolahi
Molecular and functional analysis of multimodal single-cell immune profiling data in allogeneic hematopoietic stem cell transplantation and immune checkpoint therapy
Supervisors: Il-Kang Na, Dieter Beule, Benedikt Obermayer
This project aims to enhance understanding of T cell behavior in immunological treatments like allogeneic hematopoietic stem cell transplantation (alloHSCT) and immune checkpoint inhibitors, particularly in managing adverse reactions such as graft-vs-host disease (GVHD) and immune-related adverse events (irAE). The research involves single-cell sequencing of gene expression, T cell receptor (TCR) and B cell receptor (BCR) repertoires from donor-recipient pairs before and after alloHSCT, combined with computational analysis to identify molecular patterns and variations. The project also focuses on developing computational models to predict TCR antigen specificity and correlate clonal identity with gene expression, leveraging existing databases and high-throughput data. Additionally, the study will analyze antigen specificities in T cells from irAE patients, aiming to predict and experimentally validate specific TCR-peptide interactions relevant to autoimmunity or tumor responses, ultimately improving therapeutic strategies.
Nora Koreuber
Explainable AI and Vision Transformers for Digital Pathology
Supervisors: Dagmar Kainmüller, Christoph Lippert
Explainability in biomedical image analysis is crucial for building trust in AI models used in both clinical and research contexts. My PhD project focuses on developing, evaluating, and applying explainability methods for deep learning models in digital pathology, with an emphasis on Vision Transformers. As part of this work, I have contributed to the publication of machine-learning-ready image datasets, including synthetic H&E images with controllable noise and a large multiplex immunofluorescence dataset from colon carcinoma tissue microarrays, which we use to benchmark pathology foundation models for cell phenotyping. I investigate how the architectural properties of Vision Transformers can be leveraged for interpretability and systematically review methods for vision transformer explainability. Overall, my research aims to deepen the understanding of how deep learning models make predictions on digital pathology imaging data and how this understanding can be used for new scientific insights.
Jakub Liu
Relevance-based Supervised Detection of Somatic Variants
Supervisors: Dieter Beule, Christine Sers, Ulf Leser
Susmita Mandal
Spatial Analysis of Extrachromosomal DNA in Neuroblastoma Tumors
Supervisors: Teresa Krieger, Anton Henssen
Javier Marchena Hurtado
Neighbor-based normalization for protein data in CITE-seq
Supervisors: Nils Blüthgen, Stefanie Grosswendt
CITE-seq is a method to perform RNA sequencing along with gaining quantitative information about surface proteins on a single-cell level. Nevertheless, CITE-seq data presents several sources of technical noise. There is therefore a need to normalize CITE-seq data in order to remove technical noise as much as possible. I focus on normalizing the protein data in CITE-seq. Specifically, I focus on reducing technical noise in protein data caused by different sequencing depths (similar cells occasionally have different total numbers of protein counts, due to differences in the sequencing process). Here, I propose a per-cell normalization method for CITE-seq protein data where I adjust the total protein counts of each cell, thereby reducing noise due to different sequencing depths.
Sofya Marchenko
Towards overcoming endocrine resistance in breast cancer
Supervisors: Christine Sers, Nils Blüthgen
Although endocrine therapy to block the ER pathway is highly effective and continues to be the mainstay for ER+ Breast Cancer (BC) patients, de novo resistance is common. In parallel to hormone-dependent growth in BC, tumor cells are known to heavily depend on growth factor receptor mediated signaling to execute their pathological behaviors. My aim is to gain insight into the drivers of endocrine therapy resistance using computational methods, such as machine learning and network analysis. Making use of multiple layers of omics data and integrating them with clinical data, I hope to find molecular signatures to elucidate the escape pathways, which provide tumors with alternative proliferative and survival stimuli. Due to the complexity of each patient case, understanding the activated pathways in endocrine resistant patients will help to identify those patients whose tumors are most likely to benefit from specific cotargeting strategies, which is an important step towards personalized therapy in BC.
Vinzenz May
Long read based structural variants
Supervisors: Dieter Beule
Genomic structural variants (SVs) are responsible for a wide range of rare diseases and play an important role in many cancers. A structural variant is a re-arrangement of DNA in a chromosome, e.g. duplication of a larger region or translocation from one chromosome to another. They can have direct or regulatory influence on the expression of genes and splicing variants. SVs are still difficult to correctly detect and genotype because of the limited span of basepairs per read. Such reads are generated with widely used technologies such as short read sequencing (< 300 bp). Longer reads (> 10kb) have been much more error-prone in the past but become similarly precise as short reads in resolving each single base correctly. To facilitate a broad usage of long read sequencing in clinical diagnostics, we still need to develop software to correctly find and genotype SVs in families or groups of patients or sub-clones of cancer tumors. My Project aims to develop a new pipeline and new algorithms to find all SVs in family genomes or tumor genomes so that researchers have a better and reliable tool at their hands to learn about the roles of SVs in many rare diseases and cancer types.
Cedric Moris
Multi–omic tumor evolution under treatment and environmental selective pressure
Supervisor: Dieter Beule
Therapy resistance in cancer medicine is largely driven by tumor heterogeneity, plasticity, and evolution. My PhD project focuses on the development of new computational methods to track and analyze evolution based on bulk and single-cell sequencing of cancer patient samples and patient-derived 3D tumoroids. I aim to translate findings into practical lab applications aimed at preventing or circumventing the evolution of resistance mechanisms.
Amos Münch
Supervisors: Livius Penter
Tino Petrov
Essential regulatory elements of high-level amplicons as cancer type specific growth dependencies
Supervisor: Dieter Beule, Anton Henssen
Cancer genome reconstruction and tissue-specific enhancer calling based on ONT WGS data.
Jennifer von Schlichting
Modeling the Tumor Microenvironment in Patient-Derived Organoid Culture
Supervisors: Nils Blüthgen, Chris Sander, Markus Morkel
Patient-derived organoids (PDOs) are a model of choice to elucidate inter- and intratumoral heterogeneity to combat therapy resistance. However, the utility of PDOs is limited by heterologous and poorly-defined extracellular matrices and lack of proper tumor microenvironment, thus failing to model the tumor in its complexity. I am working on developing an approach to identify relevant paracrine interactions between stromal and tumor cells in colorectal cancer (CRC). Single cell-RNAseq data of 12 CRC patients were analyzed for ligand-receptor pairs enabling stroma-to-tumor signaling. Further, I aim to model ECM composition in colorectal cancer, by supplementing the laminin/collagen IV rich environment with other known ECM proteins such as collagen I to identify the impact of a changing substrate on cell plasticity.
In an appropriate in vitro assay, I assess physiological relevance based on single cell analysis. Thus, I am trying to identify paracrine factors and signals affecting proliferation, differentiation, and developmental trajectories of CRC PDOs in vitro. I hypothesize that environmental factors may limit the phenotypic space in which organoid cells differentiate, disabling the study of more invasive behaviors in vitro. Recent data show that ECM parameters have a strong impact on cell plasticity and highlight the importance of adjusting and expanding organoid in vitro culture models. It is my goal to provide guidelines to improve existing PDO CRC models and provide a feasible approach to address common limitations in organoid culture. Ultimately, I am interested in identifying factors that can interfere with drug efficacy and potentially favor clinically relevant therapy resistance mechanisms.
Katharina Schneider
Targeted phosphoproteomics-based prediction of drug responses to combinatorial treatment
Supervisors: Nils Blüthgen
While initial treatment for high-risk Neuroblastoma (HR NB) can result in remission, it is frequently followed by lethal relapse. Therapy resistance is driven by tumor evolution, with early metastasis, genetic mutations, and dynamic phenotypic plasticity as driving factors. I am interested in classifying HR NB into signalling phenotypes based on mutational profiles. Ultimately, I will examine how these signalling phenotypes respond to both single and combinatorial drug treatments. To achieve this, mutational profiles from patient samples were extracted to generate a HR NB oncogene library. This approach will provide single-cell resolution of signaling shifts induced by various oncogenes and subsequently, drug treatments.
Oğuz Şerbetci
Explainable information extraction for patient similarity in precision oncology
Supervisors: Ulf Leser
Ani Shubitidze
The Influence of Tumor Mechanics onto Signaling Pathways Driving Proliferation and Therapy Resistance in PDAC
Supervisors: Christine Sers, Nils Blüthgen, Naveed Ishaque
Description: The goal of my project is to understand how the physical and mechanical properties of the tumor microenvironment influence the progression of pancreatic ductal adenocarcinoma (PDAC). Specifically, I am investigating how tissue stiffness and the organisation of the extracellular matrix influence tumor behavior and response to therapies. To investigate this, I apply advanced bioinformatics approaches to integrate single-cell and spatial transcriptomic data with histopathological imaging, enabling the mapping of gene expression within its spatial and structural context.
Eva Thielecke
Single Cell Data Analysis with Interpretable AI to Understand Transcriptomic States and Changes in Colorectal cancer
Supervisors: Nils Blüthgen
Karina Tristan
Generating a colorectal cancer specific oncogenic stress signature by integrating transcriptomic(bulk/single-cell) and proteomic datasets (DRAFT)
Supervisors: Christine Sers, Nils Blüthgen
Colorectal Cancer(CRC) is one of the main causes of death by cancer worldwide. It is a highly heterogeneous cancer that can be caused by several mutations the more known to be KRAS and BRAF. BRAFmut cancer is recognized as the most aggressive and has a tendency of high recurrence after treatment. Oncogenic stress signatures aimed for the classification and recognition of BRAFmut CRC are still unknown. Several problems arise to address the question. We first need to determine what is the "type" of stress and at which level(RNA/protein) are to be discussed, such as replication stress or oxidative stress. Secondly, is this stress signature able to determine the sensitivity of the cells? For these questions, a compilation of stress signatures is to be collected and applied on RNAseq data to generate a new signature specific for CRC BRAFmut, this signature then will be tried out as a biomarker after gene scoring is done. After the signature is generated it needs to be tested to check whether a correlation or difference is seen by cellular phenotype, cell/tissue types and cell states. This is to be approached by the analysis of single-cell data. Preliminary analysis with a "lab-made" signature has been done in publicly available data with rather inconclusive results, the use of other signatures for comparison is to be done as well as the use of different datasets. Gene scoring methods are being analyzed for the different signatures collected for further analysis.
Xing Wang
Question Answering for Precision Oncology
Supervisors: Ulf Leser, Il-Kang Na
Personalized therapies become increasingly available for cancer treatment and offer important alternatives especially where standard therapies fail. For a specific genomic profile of a patient, often, an extensive literature research has to be conducted to find a possible treatment. In this project, we want to facilitate the literature review process by employing text mining and machine learning methods to gather the desired information automatically from the biomedical literature. The model is designed to directly answer questions posed by clinicians and provide the corresponding literature references with it.
Tzu-Ting Wei
Analysing tumour heterogeneity in single-cell transcriptomics using somatic variants
Supervisors: Dieter Beule, Nils Bluethgen, Christine Sers
Single-cell techniques enable detailed characterisation of tumour heterogeneity. Our recent study suggested that normal cells adjacent to tumour cells exhibit a more stem-like phenotype, possibly triggered by the tumour microenvironment. However, reliable identification of tumour cells is often challenging. It is feasible to differentiate tumour cells from normal cells in copy number altered tumours via copy number variation (CNV) inference methods from transcriptomes. In contrast, copy number-based methods can not work with tumours which exhibit more single nucleotide variants (SNV) but few copy number changes. Somatic variant calling methods developed from bulk RNA sequencing data have been applied to single-cell data, while all methods suffer from the precision and recall trade-off. In this project, I will use whole-genome and whole-exome genomics data and single-cell transcriptomes to investigate the non-random distribution of tumour-associated SNVs.
Robin Xu
Mutational landscape on extrachromosomal circular DNA (ecDNA)
Supervisors: Anton G. Henssen, Leif S. Ludwig, Frederik Damm, Kerstin Haase
Erika Zuljan
Comprehensive analysis of the tumor microenvironment in advanced salivary gland cancers
Supervisors: Dieter Beule, Damian Tobias Rieke, Stefanie Großwendt
Salivary gland cancers (SGC) are rare malignancies that arise from the tissue of salivary glands and account for 5% of all head and neck cancers. The second most common histology is ACC (Adenoid Cystic Carcinoma), with a very poor outcome due to the lack of established therapy options. Compared to other entities, ACC is defined as immune-deserted. Exome (WES), genome (WGS) and bulk RNA-seq from patients with advanced ACC and other entities were analyzed to gain insights in the mutational landscape, transcriptome and tumor immune microenvironment (TIME). The cohort of 97 patients is one of the largest cohorts in the world for advanced salivary gland cancer. Preliminary result show a lower immune-infiltration in most of the ACC tumors, with the exception of small subset. Single cell data is being produced and analyzed to characterize the immune-activated ACCs and the TIME of ACCs compared to other entities. The aim here is to find potential targets for therapy for advanced ACC,with a focus on immuno-therapy.
