Cell-type deconvolution in epigenome-wide association studies: a review and recommendations

Figures & data

Figure 1.  Reference-based versus reference-free inference.

Summary of the main advantages and disadvantages of the two major inference-paradigms for adjusting cell-type heterogeneity in EWAS.

DNAm: DNA methylation; EWAS: Epigenome-wide association study.

Figure 2.  Guideline flowchart for selecting optimal cell-type deconvolution method.

Recommended guideline for selecting an optimal algorithm for cell-type deconvolution in an EWAS. In the final column, within each box, recommended algorithms have been ordered in order of preference.

COMBAT: Combatting Batch Effects; CP: Constrained projection; DNAm: DNA methylation; EWAS: Epigenome-wide association studies; ISVA: Independent surrogate variable analysis; QP: Quadratic programming; RPC: Robust partial correlation; RUV: Removing unwanted variation; SVA: Surrogate variable analysis.

Table 1.  Algorithm name, whether it is reference-free, reference-based or semireference free, what inference is possible with the algorithm, tissue types on which it has been successfully applied, whether algorithm adjusts for confounders other than cell-type composition, programming language in which it is available, website link and main reference.

Statistical algorithm	Inference paradigm	Output	Tissues successfully applied	Adjusts for other confounders?	Availability	Website	Ref.
Houseman’s CP/QP	Reference based	Cell-type fractions + DMPs	Whole blood, PBMC, cord blood, breast	No	R	https://cran.r-project.org/web/packages/RefFreeEWAS/index.html	[26]
CIBERSORT	Reference based	Cell-type fractions + DMPs	Whole blood, PBMC, breast	No	Web-based tool, JAVA, R	https://cibersort.stanford.edu/	[32,33]
RPC	Reference based	Cell-type fractions + DMPs	Whole blood, PBMC, breast	No	R	https://github.com/sjczheng/EpiDISH	[32,33]
RefFreeEWAS	Reference free, supervised	DMPs	Any	Yes, if other confounders carry high variance.	R	https://cran.r-project.org/web/packages/RefFreeEWAS/index.html	[34]
EWASher	Reference free	DMPs	Whole blood	Yes, if other confounders carry high variance.	Python, R	www.microsoft.com/en-us/download/details.aspx?id=52501 (only R version available online)	[35]
ReFACTor	Reference free	DMPs	Whole blood	Yes, if other confounders carry high variance.	Python, R	www.cs.tau.ac.il/∼heran/cozygene/software/refactor.html	[36]
SVA	Reference free, supervised	DMPs	Any	Yes	R	https://bioconductor.org/packages/release/bioc/html/sva.html	[37,38]
ISVA	Reference free, supervised	DMPs	Any	Yes	R	https://cran.r-project.org/web/packages/isva/index.html	[39]
RUV	Semireference free	DMPs	Whole blood, PBMC	No, unless control set is modified.	R	https://cran.r-project.org/web/packages/ruv/index.html	[40]
NMF–CP/QP	Reference free	Cell-type fractions + DMPs	Any	Not fully assessed	R	https://cran.r-project.org/web/packages/RefFreeEWAS/index.html	[41]

CP: Constrained projection; DMP: Differentially methylated position; ISVA: Independent surrogate variable analysis; NMF: Non-negative matrix factorization; PBMC: Peripheral blood mononuclear cell; QP: Quadratic programming; RPC: Robust partial correlation; RUV: Removing unwanted variation; SVA: Surrogate variable analysis.

Houseman EA , AccomandoWP , KoestlerDCet al. DNA methylation arrays as surrogate measures of cell mixture distribution . BMC Bioinformatics13 , 86 ( 2012 ).

 PubMed Web of Science ®Google Scholar

Newman AM , LiuCL , GreenMRet al. Robust enumeration of cell subsets from tissue expression profiles . Nat. Methods12 ( 5 ), 453 – 457 ( 2015 ).

 PubMed Web of Science ®Google Scholar

Teschendorff AE . A comparison of reference-based algorithms for correcting cell type heterogeneity in epigenome-wide association studies . BMC Bioinformatics18 ( 1 ), 105 ( 2017 ).

 PubMedGoogle Scholar

Houseman EA , MolitorJ , MarsitCJ . Reference-free cell mixture adjustments in analysis of DNA methylation data . Bioinformatics30 ( 10 ), 1431 – 1439 ( 2014 ).

 PubMed Web of Science ®Google Scholar

Zou J , LippertC , HeckermanD , AryeeM , ListgartenJ . Epigenome-wide association studies without the need for cell type composition . Nat. Methods11 ( 3 ), 309 – 311 ( 2014 ).

 PubMed Web of Science ®Google Scholar

Rahmani E , ZaitlenN , BaranYet al. Sparse PCA corrects for cell type heterogeneity in epigenome-wide association studies . Nat. Methods13 ( 5 ), 443 – 445 ( 2016 ).

 PubMed Web of Science ®Google Scholar

Leek JT , StoreyJD . Capturing heterogeneity in gene expression studies by surrogate variable analysis . PLoS Genet.3 ( 9 ), 1724 – 1735 ( 2007 ).

 PubMed Web of Science ®Google Scholar

Leek JT , JohnsonWE , ParkerHS , JaffeAE , StoreyJD . The sva package for removing batch effects and other unwanted variation in high-throughput experiments . Bioinformatics28 ( 6 ), 882 – 883 ( 2012 ).

 PubMed Web of Science ®Google Scholar

Teschendorff AE , ZhuangJ , WidschwendterM . Independent surrogate variable analysis to deconvolve confounding factors in large-scale microarray profiling studies . Bioinformatics27 ( 11 ), 1496 – 1505 ( 2011 ).

 PubMed Web of Science ®Google Scholar

Gagnon-Bartsch JA , SpeedTP . Using control genes to correct for unwanted variation in microarray data . Biostatistics13 ( 3 ), 539 – 552 ( 2012 ).

 PubMed Web of Science ®Google Scholar

Houseman EA , KileML , ChristianiDC , InceTA , KelseyKT , MarsitCJ . Reference-free deconvolution of DNA methylation data and mediation by cell composition effects . BMC Bioinformatics17 , 259 ( 2016 ).

 PubMed Web of Science ®Google Scholar