Project Details
Projekt
Digital Tissue Deconvolution - Learning from single cells
Subject Area
Medical Informatics and Medical Bioinformatics
Bioinformatics and Theoretical Biology
Bioinformatics and Theoretical Biology
Term
from 2019 to 2024
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 420069742
Digital Tissue Deconvolution (DTD) addresses the following computational problem: Given a bulk expression profile of a tissue that consists of multiple cell types such as tumor cells, lymphocytes, endothelial cells, or macrophages, what are the abundances of these cells in the tissue? For DTD we must look at the right set of marker genes. Most importantly, genes whose expression differs between tissue and reference must be excluded from analysis. Which are those genes?It is our firm believe that the full power of a learning from single-cell data approach will only unfold once large single-cell data sets become available for training. This will open new avenues in DTD including the estimation of rare cell populations, distinguishing phenotypically similar cells, and adapting DTD to specific applications, such as the deconvolution of tumor tissue.Our goals of this research proposal are (A) to use available and upcoming single-cell RNA sequencing data to train DTD loss-functions for improved deconvolution. These loss functions will be adapted to different tissue types. This will allow us to estimate rare cell populations and to distinguish phenotypically similar cells.(B) to apply DTD retrospectively on available bulk gene-expression data of tumor entities. This will reveal associations between cell compositions and clinical properties such as survival or response to treatment.(C) to provide user-friendly software that allows to apply pre-trained DTD loss-functions on bulk gene-expression data. This software will work independently from the measurement platform.(D) to provide user-friendly software for loss-function learning. This allows the user to adapt DTD to specific tissues.
DFG Programme
Research Grants
