sNucConv: A bulk RNA-seq deconvolution method trained on single-nucleus RNA-seq data to estimate cell-type...

Area of Science:

• Computational biology
• Genomics
• Biotechnology

Background:

• Deconvolution algorithms typically use single-cell RNA sequencing (scRNA-seq) on bulk RNA sequencing (bulk RNA-seq) for cell-type composition estimation.
• Adipose tissue's cellular makeup is variable, and adipocytes require single-nucleus RNA sequencing (snRNA-seq) for capture.

Purpose of the Study:

• Develop sNucConv, a deep-learning deconvolution tool utilizing snRNA-seq data.
• Validate the tool's accuracy in estimating cell-type proportions in human visceral (hVAT) and subcutaneous (hSAT) adipose tissues.
• Demonstrate the feasibility of validated deconvolution models for tissues not amenable to scRNA-seq.

Main Methods:

• Trained sNucConv using human adipose tissue snRNA-seq data.
• Corrected training data using correlated genes between snRNA-seq and bulk RNA-seq, and cell-type-specific regression models.
• Applied sNucConv to bulk RNA-seq data for cell-type proportion estimation.

Main Results:

• sNucConv estimated cell-type proportions with high accuracy (R=0.93 for hVAT, R=0.95 for hSAT).
• The model achieved high accuracy on independent validation samples.
• The developed model was depot-specific, reflecting distinct gene expression patterns in hVAT and hSAT.

Conclusions:

• sNucConv accurately deconvolutes cell-type composition in adipose tissue using snRNA-seq data.
• The tool provides depot-specific models, capturing biological differences.
• sNucConv offers a validated approach for deconvolution in tissues challenging for scRNA-seq.