Identification of novel cellular states and therapeutic targets in PDAC with machine learning
Chengxin Yu Lunenfeld-Tanenbaum Research Institute; University of of Toronto
Abstract
The 8% 5-year survival rate of pancreatic ductal adenocarcinoma (PDAC) leads to an urgent need for novel therapies. An attractive target is tumour associated stromal cells (TAS) that are involved in PDAC progression and immunosuppression. However, an exact atlas of TAS states in PDAC and their interactions with tumour cells that would identify novel drug targets is yet to be uncovered. To address this, we have harmonized and annotated the scRNA-seq data of primary and metastasis PDAC samples from 76 patients across 4 studies. We have used unsupervised learning to extract shared transcriptional signatures to infer TAS cellular states. We linked multiple TAS transcriptional signatures to T cell exhaustion in the PDAC microenvironment and tumour subtypes. We have identified novel TAS transcriptional markers and spatially mapped cellular states using imaging mass cytometry (IMC), and untangled interactions between the tumour and its stroma that plays important roles in tumour progression and maintenance. Our systematic approach to map pan-cohort TAS transcriptional programs in PDAC at single-cell resolution provides a robust subtyping scheme and a semi-automated pipeline to generate rich outputs to guide future study in anti-tumour immunity and drug development.
Keywords: Pancreatic cancer,cancer transcriptomics,cancer proteomics,microenvironment,therapeutic targets,single-cell,imaging mass cytometry,machine learning.