An informatics framework for single-cell multi-omics from clinical specimens

点击翻译按钮获取中文摘要

PROJECT SUMMARY Intra-tumor heterogeneity is a significant barrier to precision oncology. Emerging single-cell and spatial profiling approaches have enabled basic research into tumor heterogeneity. However, the application of these emerging approaches to the clinical decision process is limited. There is a critical need for predictive models that integrate these novel data with existing genomics approaches and histology, to generate actionable clinical recommendations. This proposal builds on my lab’s recent work, using single-cell RNA sequencing (scRNA-seq) to map the cellular hierarchies of complex tumors. Our preliminary data extend these studies to single-cell multi-omics, integrating single-cell assay for transposase-accessible chromatin (scATAC-seq) and spatial transcriptomics (ST). Our long-term goal is to develop models of malignant progression based on sequencing data from patient biopsies and deploy them to support clinical decisions. The overall objective of this project is to develop algorithms to integrate heterogeneous single-cell and imaging data to support therapy selection, trained on data from multiple cancers and broadly applicable pan-cancer. The rationale for this work is that these algorithms will be applied to pre-treatment biopsies to predict progression and to recommend appropriate therapy combinations. In Aim 1 we will develop and validate algorithms to model clonal composition, phylogeny, and evolutionary trajectory. This will be used to rigorously identify combinatorial chemotherapy targets and monitor emerging treatment-resistant clones. In Aim 2, we integrate scRNA-seq with ST as training data to develop a predictive model of gene expression and cellular composition, based on imaging data alone. We validate these algorithms internally, on prospective cohorts, and in situ in adjacent tissue. In Aim 3, we develop predictive models of two clinical problems that are challenging in many cancers: 1) the response to ionizing radiation, 2) the emergence of hypermutation at recurrence. Here, we exploit modern deep-and-wide learning approaches to identify genomic predictors of outcome that are tailored to a patient’s clinical context. We will validate this approach using both internal and external controls. Algorithms will be implemented in clinician dashboards in an existing system and the evaluation of clinical support will take place at two sites: the University of California, San Francisco and the University of Pittsburgh. We anticipate that this project will identify novel prognostic signatures, enable risk stratification, disease monitoring, and the selection of precision therapies. These studies will significantly advance our ability to apply single-cell and spatial profiling in the clinical setting.