Deconvolution of intra-tumoural heterogeneity using whole section images in pancreatic cancer

点击翻译按钮获取中文摘要

Keywords: Machine learning, intratumoural heterogeneity, pancreatic cancer, subtyping, artificial intelligenceMolecular subtyping is a methodology to subclassify cancer that looks otherwise very similar under the microscope and is essential to better define molecular taxonomy of prognosis and treatment response. Subtyping using mRNA expression (transcriptome) is a popular method, as it not only gives a global overview of gene expression to allow (un)supervised clustering, it also provides runways to discovery of carcinogenesis pathways that underpins each subtype. In general, RNA is extracted from a small piece of tissue of a few millimetres size, prior to transcriptome sequencing. Therefore, it does not take the whole tumour or intra-tumoural heterogeneity (ITH) into consideration.While ITH has been shown to be pervasive across cancer types, characterising them at the whole tumour scale remains technically and financially challenging. Hence, a novel solution to better understand ITH and transcriptomic subtyping is urgently needed to personalise and improve treatments for pancreatic cancer patients. This project will develop a novel strategy that leverages advanced artificial intelligence (AI) models to construct spatially resolved transcriptomic profiles of large pancreatic cancer patient cohorts in silico. It will use cutting edge deep learning techniques in computer vision and medical image analysis, bioinformatics and transcriptome analysis, and will leverage the availability of well-characterised pancreatic cancer cohorts (ICGC, TCGA and data from the national Precision-Panc trial).