AdRoit: an accurate and robust method to infer complex transcriptome composition

Abstract

RNA sequencing technology promises an unprecedented opportunity in learning disease mechanisms and discovering new treatment targets. Recent spatial transcriptomics methods further enable the transcriptome profiling at spatially resolved spots in a tissue section. In controlled experiments, it is often of immense importance to know the cell composition in different samples. Understanding the cell type content in each tissue spot is also crucial to the spatial transcriptome data interpretation. Though single cell RNA-seq has the power to reveal cell type composition and expression heterogeneity in different cells, it remains costly and sometimes infeasible when live cells cannot be obtained or sufficiently dissociated. To computationally resolve the cell composition in RNA-seq data of mixed cells, we present AdRoit, an accurate androbust method to infer transcriptome composition. The method estimates the proportions of each cell type in the compound RNA-seq data using known single cell data of relevant cell types. It uniquely uses an adaptive learning approach to correct the bias gene-wise due to the difference in sequencing techniques. AdRoit also utilizes cell type specific genes while control their cross-sample variability. Our systematic benchmarking, spanning from simple to complex tissues, shows that AdRoit has superior sensitivity and specificity compared to other existing methods. Its performance holds for multiple single cell and compound RNA-seq platforms. In addition, AdRoit is computationally efficient and runs one to two orders of magnitude faster than some of the state-of-the-art methods.

Estimating cell-type-specific gene co-expression networks from bulk gene expression data with an application to Alzheimer’s disease

“…, m} for a positive integer m. We shall assume the cell-type proportions ⇡ ik 's in (2) are given in the ensuing development, and later demonstrate in experiments that our method is not sensitive to uncertainties in ⇡ ik 's; see discussions in Section 6. To infer ⇡ ik 's from bulk samples, many methods have been developed that utilize cell type marker genes (i.e., genes that are only highly expressed in one cell type of interest) with expressions profiles gathered from pure cell types (Newman et al, 2015;Li et al, 2016) or single cell RNA-seq data (Wang et al, 2019;Newman et al, 2019;Jew et al, 2020;Dong et al, 2021;Yang et al, 2021). In these methods, the proportions ⇡ ik 's are estimated by, for example, nonnegative least squares (Wang et al, 2019) or support vector regression (Newman et al, 2019).…”

“…Using the bulk data, various methods are available to infer mean expression levels in each cell type (Newman et al, 2019), or to infer cell type proportions (Abbas et al, 2009;Wang et al, 2019;Newman et al, 2019;Tang et al, 2020;Jew et al, 2020;Yang et al, 2021). More recently, methods have been proposed to infer cell-type-specific expressions in each sample, such as CIBERSORTx (Newman et al, 2019) and bMIND (Wang et al, 2021a).…”