--- title: "FlowSorted.Blood.EPIC" author: "Lucas A Salas" date: "`r Sys.Date()`" output: rmarkdown::html_vignette vignette: > %\VignetteIndexEntry{FlowSorted.Blood.EPIC} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- ```{r setup, include = FALSE} knitr::opts_chunk$set( collapse = TRUE, comment = "#>" ) ``` Illumina Human Methylation data from EPIC on immunomagnetic sorted adult blood cell populations. The FlowSorted.Blood.EPIC package contains Illumina HumanMethylationEPIC (EPIC)) DNA methylation microarray data from the immunomethylomics group (manuscript submitted), consisting of 37 magnetic sorted blood cell references and 12 samples, formatted as an RGChannelSet object for integration and normalization using most of the existing Bioconductor packages. This package contains data similar to the FlowSorted.Blood.450k package consisting of data from peripheral blood samples generated from adult men and women. However, when using the newer EPIC microarray minfi estimates of cell type composition using the FlowSorted.Blood.450k package are less precise compared to actual cell counts. Hence, this package consists of appropriate data for deconvolution of adult blood samples used in for example EWAS relying in the newer EPIC technology. Researchers may find this package useful as these samples represent different cellular populations ( T lymphocytes (CD4+ and CD8+), B cells (CD19+), monocytes (CD14+), NK cells (CD56+) and Neutrophils of cell sorted blood generated with high purity estimates. As a test of accuracy 12 experimental mixtures were reconstructed using fixed amounts of DNA from purified cells. **Objects included:** 1. *FlowSorted.Blood.EPIC* is the RGChannelSet object containing the reference library ```{r eval=TRUE} library(FlowSorted.Blood.EPIC) if (memory.limit()>8000){ FlowSorted.Blood.EPIC <- libraryDataGet('FlowSorted.Blood.EPIC') FlowSorted.Blood.EPIC } ``` The raw dataset is hosted in both ExperimentHub (EH1136) and GEO [GSE110554](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE110554) 2. *IDOLOptimizedCpGs* the IDOL L-DMR library for EPIC arrays ```{r eval=FALSE} data("IDOLOptimizedCpGs") head(IDOLOptimizedCpGs) ``` 3. *IDOLOptimizedCpGs450klegacy* the IDOL L-DMR library for legacy 450k arrays ```{r eval=FALSE} data("IDOLOptimizedCpGs450klegacy") head(IDOLOptimizedCpGs450klegacy) ``` *See the object help files for additional information* **estimateCellCounts2 function for cell type deconvolution:** We offer the function estimateCellCounts2 a modification of the popular estimatesCellCounts function in minfi. Our function corrected small glitches when dealing with combining the DataFrame objects with the reference libraries. We allow the use of MethylSet objects such as those from GEO. However, we offer only Quantile normalization for those datasets (assuming that they have not been previously normalized). The estimates are calculated using the CP/QP method described in Houseman et al. 2012. and adapted in minfi. CIBERSORT and RPC are allowed using external packages. *see ?estimateCellCounts2 for details* ```{r eval=TRUE} # Step 1: Load the reference library to extract the artificial mixtures if (memory.limit()>8000){ FlowSorted.Blood.EPIC <- libraryDataGet('FlowSorted.Blood.EPIC') FlowSorted.Blood.EPIC } FlowSorted.Blood.EPIC library(minfi) # Note: If your machine does not allow access to internet you can download # and save the file. Then load it manually into the R environment # Step 2 separate the reference from the testing dataset if you want to run # examples for estimations for this function example RGsetTargets <- FlowSorted.Blood.EPIC[, FlowSorted.Blood.EPIC$CellType == "MIX"] sampleNames(RGsetTargets) <- paste(RGsetTargets$CellType, seq_len(dim(RGsetTargets)[2]), sep = "_") RGsetTargets # Step 3: use your favorite package for deconvolution. # Deconvolute a target data set consisting of EPIC DNA methylation # data profiled in blood, using your prefered method. # You can use our IDOL optimized DMR library for the EPIC array. This object # contains a vector of length 450 consisting of the IDs of the IDOL optimized # CpG probes. These CpGs are used as the backbone for deconvolution and were # selected because their methylation signature differs across the six normal # leukocyte subtypes. Use the option "IDOL" head (IDOLOptimizedCpGs) # If you need to deconvolute a 450k legacy dataset use # IDOLOptimizedCpGs450klegacy instead # We recommend using Noob processMethod = "preprocessNoob" in minfi for the # target and reference datasets. # Cell types included are "CD8T", "CD4T", "NK", "Bcell", "Mono", "Neu" # To use the IDOL optimized list of CpGs (IDOLOptimizedCpGs) use # estimateCellCounts2 an adaptation of the popular estimateCellCounts in # minfi. This function also allows including customized reference arrays. # Do not run with limited RAM the normalization step requires a big amount # of memory resources if (memory.limit()>8000){ propEPIC<-estimateCellCounts2(RGsetTargets, compositeCellType = "Blood", processMethod = "preprocessNoob", probeSelect = "IDOL", cellTypes = c("CD8T", "CD4T", "NK", "Bcell", "Mono", "Neu")) print(head(propEPIC$prop)) percEPIC<-round(propEPIC$prop*100,1) } ``` # Advanced user deconvolution CP/QP, CIBERSORT and/or RPC deconvolution ```{r eval=TRUE} noobset<- preprocessNoob(RGsetTargets) #or from estimateCellCounts2 returnAll=TRUE if (memory.limit()>8000){ propEPIC<-projectCellType_CP ( getBeta(noobset)[IDOLOptimizedCpGs,], IDOLOptimizedCpGs.compTable, contrastWBC=NULL, nonnegative=TRUE, lessThanOne=FALSE) print(head(propEPIC)) percEPIC<-round(propEPIC*100,1) } # If you prefer CIBERSORT or RPC deconvolution use EpiDISH or similar # Example not to run # library(EpiDISH) # RPC <- epidish(getBeta(noobset)[IDOLOptimizedCpGs,], # IDOLOptimizedCpGs.compTable, method = "RPC") # RPC$estF#RPC proportion estimates # percEPICRPC<-round(RPC$estF*100,1)#percentages # # CBS <- epidish(getBeta(noobset)[IDOLOptimizedCpGs,], # IDOLOptimizedCpGs.compTable, method = "CBS") # CBS$estF#CBS proportion estimates # percEPICCBS<-round(CBS$estF*100,1)#percentages ``` Umbilical Cord Blood ```{r Umbilical Cord Blood, eval=FALSE} # UMBILICAL CORD BLOOD DECONVOLUTION library (FlowSorted.CordBloodCombined.450k) # Step 1: Load the reference library to extract the umbilical cord samples FlowSorted.CordBloodCombined.450k <- libraryDataGet('FlowSorted.CordBloodCombined.450k') FlowSorted.CordBloodCombined.450k # Step 2 separate the reference from the testing dataset if you want to run # examples for estimations for this function example RGsetTargets <- FlowSorted.CordBloodCombined.450k[, FlowSorted.CordBloodCombined.450k$CellType == "WBC"] sampleNames(RGsetTargets) <- paste(RGsetTargets$CellType, seq_len(dim(RGsetTargets)[2]), sep = "_") RGsetTargets # Step 3: use your favorite package for deconvolution. # Deconvolute a target data set consisting of 450K DNA methylation # data profiled in blood, using your prefered method. # You can use our IDOL optimized DMR library for the Cord Blood, This object # contains a vector of length 517 consisting of the IDs of the IDOL optimized # CpG probes. These CpGs are used as the backbone for deconvolution and were # selected because their methylation signature differs across the six normal # leukocyte subtypes plus the nucleated red blood cells. # We recommend using Noob processMethod = "preprocessNoob" in minfi for the # target and reference datasets. # Cell types included are "CD8T", "CD4T", "NK", "Bcell", "Mono", "Gran", # "nRBC" # To use the IDOL optimized list of CpGs (IDOLOptimizedCpGsCordBlood) use # estimateCellCounts2 from FlowSorted.Blood.EPIC. # Do not run with limited RAM the normalization step requires a big amount # of memory resources. Use the parameters as specified below for # reproducibility. # if (memory.limit()>8000){ propUCB<-estimateCellCounts2(RGsetTargets, compositeCellType = "CordBloodCombined", processMethod = "preprocessNoob", probeSelect = "IDOL", cellTypes = c("CD8T", "CD4T", "NK", "Bcell", "Mono", "Gran", "nRBC")) head(propUCB$prop) percUCB<-round(propUCB$prop*100,1) } ``` Using cell counts instead of proportions. Note: These are random numbers, not the actual cell counts of the experiment ```{r cell counts, eval=TRUE} library(FlowSorted.Blood.450k) RGsetTargets2 <- FlowSorted.Blood.450k[, FlowSorted.Blood.450k$CellType == "WBC"] sampleNames(RGsetTargets2) <- paste(RGsetTargets2$CellType, seq_len(dim(RGsetTargets2)[2]), sep = "_") RGsetTargets2 propEPIC2<-estimateCellCounts2(RGsetTargets2, compositeCellType = "Blood", processMethod = "preprocessNoob", probeSelect = "IDOL", cellTypes = c("CD8T", "CD4T", "NK", "Bcell", "Mono", "Neu"), cellcounts = rep(10000,6)) head(propEPIC2$prop) head(propEPIC2$counts) percEPIC2<-round(propEPIC2$prop*100,1) ``` Blood Extended deconvolution ```{r Blood extended deconvolution, eval=FALSE} # Blood Extended deconvolution or any external reference #please contact # Do not run library (FlowSorted.BloodExtended.EPIC) # # Step 1: Extract the mix samples FlowSorted.Blood.EPIC <- libraryDataGet('FlowSorted.Blood.EPIC') # Step 2 separate the reference from the testing dataset if you want to run # examples for estimations for this function example RGsetTargets <- FlowSorted.Blood.EPIC[, FlowSorted.Blood.EPIC$CellType == "MIX"] sampleNames(RGsetTargets) <- paste(RGsetTargets$CellType, seq_len(dim(RGsetTargets)[2]), sep = "_") RGsetTargets # Step 3: use your favorite package for deconvolution. # Deconvolute the target data set 450K or EPIC blood DNA methylation. # We recommend ONLY the IDOL method, the automatic method can lead to severe # biases. # We recommend using Noob processMethod = "preprocessNoob" in minfi for the # target and reference datasets. # Cell types included are "Bas", "Bmem", "Bnv", "CD4mem", "CD4nv", # "CD8mem", "CD8nv", "Eos", "Mono", "Neu", "NK", and "Treg" # Use estimateCellCounts2 from FlowSorted.Blood.EPIC. # Do not run with limited RAM the normalization step requires a big amount # of memory resources. Use the parameters as specified below for # reproducibility. # if (memory.limit()>8000){ prop_ext<-estimateCellCounts2(RGsetTargets, compositeCellType = "BloodExtended", processMethod = "preprocessNoob", probeSelect = "IDOL", cellTypes = c("Bas", "Bmem", "Bnv", "CD4mem", "CD4nv", "CD8mem", "CD8nv", "Eos", "Mono", "Neu", "NK", "Treg"), CustomCpGs =if(RGsetTargets@annotation[1]=="IlluminaHumanMethylationEPIC"){ IDOLOptimizedCpGsBloodExtended}else{IDOLOptimizedCpGsBloodExtended450k}) perc_ext<-round(prop_ext$prop*100,1) head(perc_ext) } ``` ```{r} sessionInfo() ``` **References** LA Salas et al. (2018). An optimized library for reference-based deconvolution of whole-blood biospecimens assayed using the Illumina HumanMethylationEPIC BeadArray. Genome Biology 19, 64. doi: [10.1186/s13059-018-1448-7](https://dx.doi.org/10.1186/s13059-018-1448-7). LA Salas et al. (2022). \emph{Enhanced cell deconvolution of peripheral blood using DNA methylation for high-resolution immune profiling}. Nat Comm (In Press). doi (biorxiv): [10.1101/2021.04.11.439377](https://dx.doi.org/10.1101/2021.04.11.439377). DC Koestler et al. (2016). Improving cell mixture deconvolution by identifying optimal DNA methylation libraries (IDOL). BMC bioinformatics. 17, 120. doi: [10.1186/s12859-016-0943-7](https://dx.doi.org/10.1186/s12859-016-0943-7). K Gervin, LA Salas et al. (2019) \emph{Systematic evaluation and validation of references and library selection methods for deconvolution of cord blood DNA methylation data}. Clin Epigenetics 11,125. doi: [10.1186/s13148-019-0717-y](https://dx.doi.org/10.1186/s13148-019-0717-y). EA Houseman et al. (2012) DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics 13, 86. doi: [10.1186/1471-2105-13-86](https://dx.doi.org/10.1186/1471-2105-13-86). [minfi](http://bioconductor.org/packages/release/bioc/html/minfi.html) Tools to analyze & visualize Illumina Infinium methylation arrays.