A sNucSeq-validated AI-based deconvolution method (sNuConv) to extract cell-type proportions of human subcutaneous and visceral adipose tissues from bulk RNA-sequencing

Adipose tissue exhibits enormous plasticity of its cellular composition, in response to weight changes and when comparing adipose tissues from different anatomical locations (depot). Deconvolution algorithms rely on single-cell RNA-sequencing data applied onto bulk-RNA-sequencing to extract information on the cell-types composition and proportions comprising a certain tissue. However, adipocytes – the functionally unique cell type of adipose tissue, are not amenable to single-cell RNA-sequencing, a challenge recently met by applying single-nucleus RNA-sequencing (sNuc-Seq). Here we aimed to develop a sNuc-Seq validated deconvolution method to extract the cellular composition from bulk RNA-sequencing of human visceral and subcutaneous adipose tissues (hVAT and hSAT, respectively). Seven hVAT and five hSAT samples, respectively, were analyzed by both bulk-RNA-sequencing and sNuc-Seq’s allowing to correlate deconvolution-predicted cell-type proportions to true proportions as assessed by sNuc-Seq. sNuc-Seq uncovered 15 in hVAT and 13 in hSAT distinct cell types. Using leave-one-out validation, existing deconvolution tools – SCDC, MuSiC, and Scaden exhibited low (mean R2<0.1 for predicted vs. true correlations) performance in uncovering cell-type proportions. Notably, prediction accuracy slightly improved when decreasing the number of cell-type groups. We therefore developed a novel, Scaden-based method - sNuConv, founded on prediction by sNuc-Bulk correlated genes and cell-type proportions correction using individual cell-type regression models. Applying sNuConv on our data resulted in cell-type proportion prediction accuracy with R2=0.81±0.13 (range:0.58–0.96) for hVAT, and R2=0.90±0.04 (range:0.85–0.96) for hSAT. We propose sNuConv as a novel, AI-based, sNuc-Seq -validated method to deduce the cellular landscape of hVAT and hSAT from bulk RNA-Seq data