I want to check if decidua tissue samples in dataset (from GEO database) GSE60438 are mixed with blood. I merged dataset GSE60438 with dataset GSE73685 consisting of multiple tissues including decidua and maternal and umbilical cord blood. I selected only healthy samples with decidua and blood from both datasets, merged their expression matrices by rownames (matrices were log transformed, quantile normalized and rownames were mapped do Entrez identifiers before that) and removed batch effect using R ComBat function with model matrix based on tissue and with datasets' accession number as batch
# secondaryaccession is a column with samples' dataset accession number: GSE60438 or GSE73685
batch = as.factor(pdata$secondaryaccession)
# Biological.Specimen is a column with tissue types in phenodata dataframe pdata
mod = model.matrix(~as.factor(Biological.Specimen), data=pdata)
# mrgd is a dataframe - expression matrices of two datasets merged by rowname
exprs = ComBat(dat=as.matrix(mrgd), batch=batch, mod=mod, par.prior=TRUE, prior.plots=FALSE)
Principal component analysis showed some of the sample leaning towards blood samples. How to check more rigorously whether those samples are really mixed with blood samples?
Сашко Лихенко There are many deconvolution algorithms published in the literature, which may be useful.
Check this review for a range of available tools. The performance of each tool depends on the underlying model assumptions, source of expression data and the availability of expression levels in known cell populations.
Article An assessment of computational methods for estimating purity...
I have used ABSOLUTE, ESTIMATE and MuTect, with each generating relevant results.
I would start by testing the expression of genes whose expression is restricted to white blood cells, e.g. MPO (myeloid) or TCR (T-Lympho), Ig-chains (B-Lympho).
- Badges
- Science topic
- Similar topics
- Engineering
- Petroleum Engineering
- Oil