Tumor Immunology and Tumor Immunotherapy

The possibilities of algorithmic approaches in the detailed analysis of tissue images are addressed in this project. Automated classification of tissue image elements can be performed in multiple ways. Systematic analysis and optimization of selected features, optimizing the algorithmic backbone (ResNet, AlexNet etc.) and training and validating the convolutional neural network (CNN). 

The analysis of colorectal cancer primary tumor cohorts and prediction of prognosis was performed and published in (Kather JN et al, 2018, PLOS Medicine). This publication was the most cited publication of the year 2019 of PLOS Medicine. This approach was then further developed into an effort to automatically identify microsatellite-instable lymphocyte-rich colorectal cancers, published in Nature Medicine (Kather JN et al, 2019, Nature Medicine). The current work with machine learning algorithms seeks to implement a platform that allows to stratify explant models and bioprints (see section 1.2.4) automatically. This will allow to systematically perturbate the model systems and automatically annotate the changes into a prognostic subgroup through machine learning. This can then be applied for identification of possible targets for intervention or to estimate the effects induced by a specific drug or intervention on the tumor microenvironment. 

 [AM1]In text citation hier einfügen?

 [HN2]Yes

The increasing number of heterogeneous datasets or limited-size datasets (as for subgroups of specific cancer entities or specific rare cancer entities) make systematic analyses challenging. Heterogeneous datasets and limited number of cases inhibit systematic approaches in a way that can annihilate efforts in precision medicine. This problem is addressed in two approaches. Topological algebra and computational topology is a framework that can better resemble the deeply interwoven relationships of biology represented in datasets (Moraleda et al, 2020, Semin Immunol, Valous et al, 2020 Semin Immunol). Therefore, a non-sql graph database was implemented in Neo4j and contains heterogeneous complex data. Now, in the next steps, systematic application of computational topology algorithms allows to classify and cluster datasets in novel ways. In addition to this, efforts have been initiated to leverage quantum computing methodology. In short, specific formulation of questions in a biquadratic layout allow for improved optimization results. Based on the quantum annealer technology, the systematic analysis of graph networks is being explored and a first prototype has been developed, allowing further optimization with geometric algebra to link the localization of identified central nodes to larger branches of the graph. This investigation of quantum methodology is a joint project together with collaboration partners from Fraunhofer IKS (Munich). The quantum processing advantage for limited datasets is further investigated for machine learning, where first results have been obtained (Matic A et al, ArXiv, 2022)[AM3] [HN4] . Currently, preparations for an international proposal are underway, on one hand investigating the optimal hardware and algorithmic approach and on the other hand investigating the utility for given clinical subsets, e.g. triple-negative breast cancer. 

 [AM1]Könnte man diese Referenzen zu den Datensätzen genauer spezifizieren, z.B. über eine Publikation?

 [HN2]Ja

 [AM3]Auch hier sollte weitestgehend das Format der in text citation genutzt werden.

 [HN4]Eingefügt

 [AM1]Könnte man diese Referenzen zu den Datensätzen genauer spezifizieren, z.B. über eine Publikation?

 [HN2]Ja

 [AM3]Auch hier sollte weitestgehend das Format der in text citation genutzt werden.

 [HN4]Eingefügt

 [AM1]Könnte man diese Referenzen zu den Datensätzen genauer spezifizieren, z.B. über eine Publikation?

 [HN2]Ja

 [AM3]Auch hier sollte weitestgehend das Format der in text citation genutzt werden.

 [HN4]Eingefügt

Classical immunotherapies focus on the modulation or activation of the adaptive arm of the immune system. Although excellent therapeutic successes have been achieved for some tumor diseases, others do not respond at all. In order to better leverage the potential of immunotherapy, we explore the role of the innate arm of the immune system. Since the microbiome composition can strongly impact therapy outcome, we focus on the interplay of microbiome and innate immune system within the tumor microenvironment. In this frame, we developed a pipeline utilizing Nanopore technology for 16S sequencing to systematically assess microbiome composition. Furthermore, we apply specific and highly sensitive PCR assays for measuring key factors impacting the immune system and therefore patient prognosis. These include qPCR assays to quantify Fusobacterium nucleatum and its products colibactin, Bacteroides fragilis toxin and FadA. Beyond the classical microbiome analysis, our research also focuses on the investigation of the mycobiome. Systematic analyses are currently being carried out to delineate fungal species on the basis of existing databases such as UNITE and WARCUP. First results of the above depicted pipeline with samples from the LUMINESCENCE trial (see section 1.2.1) show promising results with cross-entity patterns associated with either good or bad prognosis or treatment outcome. To better understand the microbiome composition within tissue, another technology has been further developed: assembloids or bioprints are artificially created tissues with a defined content of cell populations (i.e. cancer cells, T cells, macrophages, bacteria etc.). The matrix or scaffold consists of a collagen and cytokeratin containing substance, where the cells of interest are placed inside. This miniaturization allows systematic side-by-side comparison of different conditions or compositions. The effect of distinct cytokine mixtures, bacteria or bacterial communities on defined populations of tumor and immune cells is analyzed. A first use case of this approach has been published recently (Ahmed A et al, 2022, Frontiers Immunology).