Monotonic genes in results of differential gene expression analysis
1
0
Entering edit mode
@14ef1b09
Last seen 8 weeks ago
Egypt

I want to know which methods are commonly used for RNA seq data if the goal is to identify monotonic increasing or decreasing gene expression over time or across stages using the results of differential gene expression analysis?

DifferentialExpression rnaseqGene RNASeqData • 956 views
ADD COMMENT
0
Entering edit mode

Cross-posted to Biostars https://www.biostars.org/p/9585490/

ADD REPLY
3
Entering edit mode
@gordon-smyth
Last seen 2 hours ago
WEHI, Melbourne, Australia

The easiest way is to fit a time trend, select genes with a significant F-test and then check whether the fitted values are in monotonic order.

ADD COMMENT
0
Entering edit mode

Thank you so much Prof. Gordon

ADD REPLY
0
Entering edit mode

I tried to fit a time trend model for each gene but it takes forever. It seems like it is computationally expensive. Can I fit the linear model for all genes simultaneously using a matrix?

Processor: 11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz 2.80 GHz RAM:16.0 GB

  time_trend_model <- lapply(names(ordered_counts)[1:5429], function(gene) {
  formula <- as.formula(paste(ordered_counts, "~ stage"))
  lm(formula, data = ordered_counts)
  })
ADD REPLY
2
Entering edit mode

Fitting genewise time trends take only a few seconds using either limma or edgeR, even for a large dataset and even on a laptop computer. Both packages perform F-tests. There are examples in the limma and edgeR User's guides, but it is just a standard analysis for either of those packages.

I was certainly not recommending that you fit linear models directly to ordered counts. That wouldn't make any sense statistically, even it was computationally efficient.

ADD REPLY
0
Entering edit mode

trend plots

I am not sure why the plots are different from the user guide. I am dealing with defined stages while the user guide deals with hours

design = model.matrix(~0 + dge_stage$samples$Stage)
colnames(design)[1]="Stage1"
colnames(design)[2]="Stage2"
colnames(design)[3]="Stage3"
colnames(design)[4]="Stage4"

stages=as.numeric(c(dge_stage$samples$Stage))

polycurve = poly(stages, degree=3)
design <- model.matrix(~ polycurve)

dge_stage <- estimateGLMCommonDisp(dge_stage, design)

dge_stage <- estimateGLMTagwiseDisp(dge_stage, design)

glmfit=glmQLFit(dge_stage, design=design)
plotQLDisp(glmfit)

Time_FTest=glmQLFTest(glmfit, coef=ncol(glmfit$design))

#The topTags function lists the top set of genes with most 
#significant time effects.

tab <- as.data.frame(topTags(Time_FTest, n=30))

summary(decideTests(Time_FTest))

logCPM.obs <- cpm(dge_stage$counts, log=TRUE, prior.count=Time_FTest$prior.count)
logCPM.fit <- cpm(Time_FTest, log=TRUE)


par(mfrow=c(2,2))
for(i in 1:10) {
GeneID <- row.names(tab)[i]
Symbol <- tab$hgnc_symbol[i]
logCPM.obs.i <- logCPM.obs[GeneID,]
logCPM.fit.i <- logCPM.fit[GeneID,]
plot(stages, logCPM.obs.i, ylab="log-CPM", main=Symbol, pch=16)
lines(stages, logCPM.fit.i, col="green", lwd=2)
}

Also, results of summary(decideTests(Time_FTest))        

polycurve3
Down            0
NotSig       5429
Up                0
ADD REPLY

Login before adding your answer.

Traffic: 621 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6