Charting a structural and biochemical ECM niche in solid tumours Group Zlobec, Williams The extracellular matrix (ECM, matrisome) forms part of the triad of the tumour ecosystem but is understudies in terms of its contribution to tumour biology. The Williams group utilizes spatially resolved mulit-matrisomics in 2D and 3D to deeply characterize the structural and biochemical manifestations of the matrisome and the association of this to patient outcome. We work across a variety of solid tumour types including colorectal cancer and pancreatic ductal adenocarcinoma. Multi-modal assessment of extracellular matrix in solid tumours. Top: maximum intensity projection of collagen visualisation using 3D open top light sheet microscopy. Middle: spatial transcriptomics analysis of tumour budding. Bottom: digitial image analysis of fibrous structures from H&E
Digital pathology & AI to gain novel insights into colorectal cancer Group Zlobec, Williams Our Sinergia project uses AI to gain new insights into the biology of colorectal cancers. We investigate morphomolecular relationships, including the molecular subtypes and intratumoral heterogeneity in order to learn new interpretable & clinically important features from histopathology images. We use various computational methods, including graphs and deep learning) to evaluate the structural and spatial patterns at the tumor invasion front in neoadjuvantly treated patients. We’ve extended our scope to understanding transcriptional subtypes using spatial transcriptomic and spatial protein expression analysis. The tumor microenvironment, with its complex stromal patterns and immune contexture are important focus points. Collaborators on this project include M. Rodriguez (IBM Research), M. Anisimova (ZHAW), B. Snijder (ETH Zürich), A. Fischer (HES-SO & UniFribourg) and V. Koelzer (UniZürich). Epithelial cell and lymphocyte graphs in colorectal cancer
Building tools for computer-assisted diagnostics Group Zlobec, Williams In addition to exploratory tissue analysis, our team builds, tests and validates in-house, open-source and commercially available algorithms for potential diagnostic use and workflow integration. We are generating a pan-lymph node metastasis algorithm using state-of-the-art deep learning methods. We then streamline processes from the lab to data analysis, and on to visualisation of results and interaction of our algorithms with pathologists scores and feedback. By incorporating text feedback along with image predictions, vision-language models can be used to train a new model by learning from human feedback. The ultimate goal is to improve the overall performance of MetAssist, resulting in an accurate nodal screening tool to assist pathologists in their routine clinical work.Together with our expert pathologist colleagues, we collaborate on a variety of algorithms including PD-L1 (Tereza Losmanova), H. pylori (Bastian Dislich), IBD scoring (Aart Mookhoek), tumor budding- CD8 scores (Heather Dawson), breast biomarkers (Wiebke Solass) and pancreas pathology (Martin Wartenberg). Computational Analysis of Colorectal Cancer Metastases in Lymph Nodes
