The role of scale in the estimation of cell-type proportions

Abstract

Complex tissues are composed of a large number of different types of cells, each involved in a multitude of biological processes. Consequently, an important component to understanding such processes is understanding the cell-type composition of the tissues. Estimating cell type composition using high-throughput gene expression data is known as cell-type deconvolution. In this paper, we first summarize the extensive deconvolution literature by identifying a common regression-like approach to deconvolution. We call this approach the Unified Deconvolution-as-Regression (UDAR) framework. While methods that fall under this framework all use a similar model, they fit using data on different scales. Two popular scales for gene expression data are logarithmic and linear. Unfortunately, each of these scales has problems in the UDAR framework. Using log-scale gene expressions proposes a biologically implausible model and using linear-scale gene expressions will lead to statistically inefficient estimators. To overcome these problems, we propose a new approach for cell-type deconvolution that works on a hybrid of the two scales. This new approach is biologically plausible and improves statistical efficiency. We compare the hybrid approach to other methods on simulations as well as a collection of eleven real benchmark datasets. Here, we find the hybrid approach to be accurate and robust.deconvolution, gene expression, microarray, RNA-seq .

Benchmark of cellular deconvolution methods using a multi-assay reference dataset from postmortem human prefrontal cortex

“…To test the performance of different approaches to computational deconvolution of RNA-seq data (Table 1), six deconvolution methods (DWLS, Bisque, MuSiC, BayesPrism, CIBERSORTx, and hspe) were run on this DLPFC dataset of 110 bulk/fractionated RNA-seq samples [5][6][7][8][9]11] to estimate proportions of seven broad cell types. The reference snRNA-seq dataset was subset to the Mean Ratio top25 cell type marker genes for deconvolution.…”

“…Increasing numbers of bulk RNA-sequencing (RNA-seq) and single cell or nucleus RNA-seq (sc/snRNA-seq) datasets have been generated, sometimes uniformly processed, and publicly shared [1][2][3][4]. RNA-seq data historically has been cheaper to generate than sc/snRNA-seq data, leading to a surge in methods that perform cellular deconvolution and estimation of cell type proportions using reference sc/snRNA-seq data [5][6][7][8][9][10][11]. Some methods can use these estimated cell type proportions to deconvolve cell type specific gene expression [12][13][14] to overcome cellular heterogeneity and identify nuanced gene expression signals that would otherwise be masked in bulk RNA-seq data.…”

“…CIBERSORTx [11], and hspe [9] previously known as dtangle [44]. These algorithms among the best performers in recent benchmarking studies [17][18][19][20]22] and optimize predictive performance by extension of weighted least squares [5], assay bias correction [6], source bias correction [7], Bayesian methods [8], high collinearity adjustment [9], and machine learning [11].…”

Toward Precision Radiotherapy: A Nonlinear Optimization Framework and an Accelerated Machine Learning Algorithm for the Deconvolution of Tumor-Infiltrating Immune Cells

“…Several issues with the linear framework have been identified [ 30 ]. More recently, the authors of [ 36 ] showed problems associated with different scales of gene expression within the linear framework and then proposed the following hybrid model: where accounts for systemic technical variation. Equivalently, …”

How many markers are needed to robustly determine a cell’s type?

“…The role of marker genes in deconvolution remains particularly unclear: a recent benchmark suggests iScience Article that the quality of markers is more important than the deconvolution method (Avila Cobos et al, 2020), and in most studies the influence of the number of markers is only partially assessed (Newman et al, 2019;Hunt and Gagnon-Bartsch 2021). Our annotation assessment suggested that cell types are best captured with 10-200 meta-analytic markers; deconvolution is a natural place to test this heuristic.…”