TOAST: improving reference-free cell composition estimation by cross-cell type differential analysis

Abstract

In the analysis of high-throughput data from complex samples, cell composition is an important factor that needs to be accounted for. Except for a limited number of tissues with known pure cell type profiles, a majority of genomics and epigenetics data relies on the "reference-free deconvolution" methods to estimate cell composition. We develop a novel computational method to improve reference-free deconvolution, which iteratively searches for cell type-specific features and performs composition estimation. Simulation studies and applications to six real datasets including both DNA methylation and gene expression data demonstrate favorable performance of the proposed method.

DeCompress: tissue compartment deconvolution of targeted mRNA expression panels using compressed sensing

“…Such reference datasets are much more widely available than compartment-specific expression on the same targeted panel. We benchmarked DeCompress against reference-free methods (20,22,(24)(25)(26) using in-silico GTEx mixing experiments (53,54), 4 published datasets with known compartment proportions (11,23,58,59), and a large, heterogeneous NanoString nCounter dataset from the CBCS (43,55). In these analyses, we showed that DeCompress recapitulated true compartment proportions with the minimum error and the strongest compartment-specific positive correlations, especially when the reference dataset is properly aligned with the tissue assayed in the target.…”

“…First, DeCompress has a high computational cost, owing mainly to its lengthy compressed sensing training step. We recommend running mainly linear optimization methods in this step and have implemented parallelization options to bring computation time on par with the iterative framework proposed in TOAST (25). However, DeCompress estimates compartment proportions both accurately and precisely, compared to other reference-free methods, and provides a strong computational alternative that is much faster than costly lab-based measurement of composition.…”

“…We benchmarked DeCompress performance across 6 datasets (see Supplemental Table S2): (1) insilico mixing experiments using tissue-specific expression profiles from the Genotype-Tissue Expression (GTEx) Project (53,54), (2) expression from 4 published datasets with known compartment proportions (11,23,58,59), and (3) and tumor expression from the Carolina Breast Cancer Study (43,55). We compared the performance of DeCompress against 5 other reference-free deconvolution methods (summarized in Supplemental Table S1): deconf (20), Linseed (22), DeconICA (24), iterative non-10 negative matrix factorization with feature selection using TOAST (TOAST + NMF) (25), and…”

ADAPTS: Automated deconvolution augmentation of profiles for tissue specific cells

“…Root mean square error (RMSE) directly measures the error between the actual and predicted values for a cell type. A detailed discussion of these metrics and other alternatives are presented by Li and Wu [20].…”

Rapid preliminary purity evaluation of tumor biopsies using deep learning approach

“…Recently, various computational approaches had been developed to investigate the purities of tumor biopsies. These methods had successfully utilized the molecular signatures, such as gene expressions (e.g., the ESTIMATE algorithm [2] ), copy number aberrations (e.g., the ABSOLUTE algorithm [3] ) and DNA methylations (e.g., the LUMP algorithm [1] ), to either estimate the purity [4] , [5] , [6] , [7] , or decode the cell compositions of biopsy samples [8] , [9] , [10] , [11] , [12] , [13] , [14] . Despite the high accuracy and consistency demonstrated in these studies, however, the majority of these methods only focused on biopsy samples from the TCGA (The Cancer Genome Atlas) project [15] and very few had been validated outside the TCGA datasets.…”