Fig 1. Overview of BSCET.

The method contains two steps. (A) First, we employ the deconvolution method, such as MuSiC [14], to infer the cell type proportion, $p_j^k$, of the bulk tissue for each individual j of cell type k by incorporating the cell-type-specific expression information, ${\theta }_{jg}^k$, and cell size factor, $S_j^k$, obtained from scRNA-seq data. (B) Second, at each transcribed SNP (tSNP) site, we align the tSNP alleles with respect to its reference and alternative alleles, T and t, across individuals, and model the expression difference between two allele-specific read counts, $X_j^T$ and $X_j^t$, over the estimated cell type compositions, $p_j^k$, through a linear model with an intercept term α to capture the information not explained by the selected K cell types, where both m_j and $S_j^k$ are assumed known. The difference of population-level relative abundance between two transcribed alleles, θ^k, can therefore be inferred for each cell type. By testing for H₀: θ^k = 0, we can detect cell-type-specific allelic expression imbalance across samples.