Hello

How to get TMM normalized matrix from count matrix using edgeR

I saw that edgeR guide recommend estimateDisp directly rather estimateCommonDisp -> estimateTagwiseDisp

http://www.bioconductor.org/packages/release/bioc/vignettes/edgeR/inst/doc/edgeRUsersGuide.pdf

And I confirmed that those results are different. Also, low number of DEGs(by exactTest)

in estimateCommonDisp -> estimateTagwiseDisp.

But I don't know how to get TMM matrix from output of estimateDisp

In case of estimateCommonDisp -> estimateTagwiseDisp, I found method to get it (from below community)

https://www.biostars.org/p/317701/ by lieven.sterck

Could you explain what is right (or better) method to get TMM normalized matrix.

Below is my R code.

#### Input: Gene Symbol|Patient Index by raw count

input = read.table("Input/Liu.txt", header=TRUE, sep="\t")

rownames(input) <- input[,1]

input <- input[,-1]


#### Input Class

class <- read.table("Input/Liu Class.txt", header = TRUE, sep="\t")

library(edgeR)

group <- class$Class

design <- model.matrix(~group)

d <- DGEList(counts=input, group=group)

keep <- filterByExpr(d, group=group)

d <- d[keep, keep.lib.sizes=FALSE]

d <- calcNormFactors(d, method="TMM")

## Method1: Direct 
d1 <- estimateDisp(d, design=design) # estimateDisp=estimateCommonDisp+estimateTagwiseDisp, estimateDisp is recommended. Result little different

## Method2: Indirect

d2 <- estimateCommonDisp(d, verbose=TRUE)

d2 <- estimateTagwiseDisp(d, trend="none")



### TMM Matrix from Method2

TMM_Norm <- t(t(d2$pseudo.counts)*(d2$samples$norm.factors))


## SessionInfo
sessionInfo()
R version 3.6.0 (2019-04-26)
Platform: x86_64-redhat-linux-gnu (64-bit)
Running under: CentOS Linux 7 (Core)

Matrix products: default
BLAS/LAPACK: /usr/lib64/R/lib/libRblas.so

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8    LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] e1071_1.7-3     caret_6.0-86    ggplot2_3.3.2   lattice_0.20-38 edgeR_3.28.1    limma_3.42.2   

loaded via a namespace (and not attached):
 [1] Rcpp_1.0.5           pillar_1.4.6         compiler_3.6.0       gower_0.2.2          plyr_1.8.6          
 [6] iterators_1.0.12     class_7.3-17         tools_3.6.0          rpart_4.1-15         ipred_0.9-9         
[11] lubridate_1.7.9      lifecycle_0.2.0      tibble_3.0.3         gtable_0.3.0         nlme_3.1-149        
[16] pkgconfig_2.0.3      rlang_0.4.7          Matrix_1.2-17        foreach_1.5.0        rstudioapi_0.11     
[21] prodlim_2019.11.13   stringr_1.4.0        withr_2.2.0          dplyr_1.0.2          pROC_1.16.2         
[26] generics_0.0.2       vctrs_0.3.2          recipes_0.1.13       locfit_1.5-9.4       stats4_3.6.0        
[31] grid_3.6.0           nnet_7.3-12          tidyselect_1.1.0     data.table_1.13.0    glue_1.4.1          
[36] R6_2.4.1             survival_3.2-3       lava_1.6.7           reshape2_1.4.4       purrr_0.3.4         
[41] magrittr_1.5         ModelMetrics_1.2.2.2 scales_1.1.1         codetools_0.2-16     ellipsis_0.3.1      
[46] MASS_7.3-51.4        splines_3.6.0        timeDate_3043.102    colorspace_1.4-1     stringi_1.4.6       
[51] munsell_0.5.0        crayon_1.3.4

Use cpm, as the edgeR User's Guide says in Section 2.16.

You seem to be assuming that dispersion estimation affects the CPM matrix, but it does not.