PSEA deconvolution; some of the coefficient is NA?
0
0
Entering edit mode
kavator ▴ 20
@kavator-22955
Last seen 16 days ago
Singapore

Hi folks i tried deconvoluting with PSEA package

first, according to vignette; i defined the markers before this; none of the probesets contain any NA values;

      neuronref<-marker(neuronprobeset,expr=counts,targetMean = 100)
mgref<-marker(mgprobeset,expr=counts,targetMean = 100)

astrocyteref<-marker(astrprobeset,expr=counts,targetMean = 100)


Defined the groups (corresponding to my expression matrix as well:

    x<-c(0,1)
groups<-rep(x,each=4)


However when i defined linear model

model.group0 <- lm(counts["Apoe",] ~ neuronref + astrocyteref +
mgref,subset=which(groups==0))


I realised that mgref is NA and

summary (model.group0)


only shows Estimates of neuronref, astrocyteref. The rest; whether it is p-values or degree of freedom are NA. this problem does not happen if i remove subset=which(groups==0);

Would appreciate some advice, thank you.

PSEA Deconvolution • 364 views
0
Entering edit mode

Hi, most probably what is happening is that your dataset does not contain enough samples. Here you are trying to fit (an intercept plus) 3 population-specific markers (neuronref, astrocyteref and mgref) using 4 samples (which(groups==0)), which is like fitting a straight line across 2 data points, leaving no residual degrees of freedom (hence the problem disappearing when you suppress the "subset" option).

Just 1 more sample would (technically) allow you to fit the model but in practice, performing deconvolution will typically require dozens of samples (the exact number depend on various parameters, like e.g. the level of noise of your expression levels).