.png){kind=icon}
.png){kind=icon}
Return
Approaching the holistic transcriptome-convolution and deconvolution in transcriptomics
.png){kind=icon}
PRE
AI
DOI：10.1093/bib/bbaf388.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Tissues, organs, and entire organisms are composed of diverse cell populations, which are characterized by cell-type-specific gene activities. Bulk RNA-seq represents a robust, cost-effective, scalable method to measure gene activity at the bulk tissue level. However, pathomolecular processes lead to divergent changes in tissue composition and cell-type-specific gene deregulations, which cannot be resolved at the tissue bulk level without information on either change in cell-type proportion or expression at the single-cell level. Accordingly, methods have been developed that constrain bulk deconvolution by information from single-cell expression or cell-type proportion. In parallel, convolution methods have been developed to project single-cell expression to bulk tissue level (pseudobulk simulation). In the present review, we provide an overview of existing convolution and deconvolution methods, their interconnectivity, and benchmarking. Our unique approach lies in the joint consideration of both directions in a holistic transcriptome model. Through analysis of published (de)convolution studies and benchmarks, we identified the reduced availability of suitable datasets and the use of inaccurate convolution-like methods for (de)convolution model assessment and training as key bottlenecks in the field. On that basis, we conclude with a holistic transcriptome model envisioning that a more integral approach to convolution and deconvolution is needed. With our suggestions for a unified framework we aim to spark collaborative efforts to enable major leaps forward in the field of (de)convolution.
Keywords:
convolution
deconvolution
bulk RNA-Seq
scRNA-seq
transcriptomics
machine learning
holistic transcriptome
cell-type proportions
Journal
IF:
7.7
Papers:
5.6K
Citations:
2.7W