Differential network analysis for two treatments

Question

[John]

Hi

I have RNA seq gene expression matrix (RSEM counts) of control, positive treatment &negative treatment (two treatments, ex: knockdown, over expression)

How can I show (in single figure) some genes are unregulated in positive treatment, down regulated in negative treatment and normal in control, using any R package?

All three conditions are triplicated

Thanks in anticipation

Update: italic

[Prakash]

The simple way would be to use top 500 variable genes and make a heatmap to show genes that are variable across different conditions

[Lila M]

You can use DESeq2 for differential expression analysis (see in here ) and plot the top 100 or 500 more expressed genes. As an example:

library(DESeq2)
dds <- DESeqDataSetFromMatrix(
  countData = count.table,
  colData = samples,
  design = ~ status)
dds2 <- DESeq(dds)
res<- results(dds2)
heatmap <- assay(vsd)[head(order(res$padj), 500), ]

[John]

thank you,

assays(vsd)

gave me the variable genes between samples, I have three samples per condition. I need variable genes per group, not samples. Any idea?

Answer 1

[Kevin Blighe]

As per Lila, if you follow the DESeq2 vignette (see the link in Lila's answer), then you will learn a lot about how you can achieve what you are aiming to do.

My recommendation is to conduct a differential expression analysis between your positive and negative treatments - this will produce a table of test statistics (p-values, fold changes, etc). With this table of test statistics, you can identify the genes / variables that are statistically significantly differentially expressed between your positive and negative treatments. From this, you can obtain a list of genes that are differentially expressed.

With the list of differentially expressed genes, you can subset your matrix such that the matrix only includes these genes. Then, you can create a heatmap, which should show separation between your positive and negative treatments.

Kevin

[lalithabhuvan]

Hi Kevin,

You may be able to help me! I have the differentially expressed genes from a microarray experiment (lncRNA and mRNA) and trying to create a network of differentially expressed lncRNAs and co-expressed genes. Is it possible to do this?

Here are the list of the variables I have in the data.

ProbeName   Fold Change Regulation type seqname GeneSymbol  source  RNAlength   chrom   strand  txStart txEnd   Xhyb    probeCoordinates    Sequence    EntrezID    relationship    Associated_gene_acc Associated_gene_name    Associated_protein_name Associated_gene_strand  Associated_gene_start   Associated_gene_end [KO](raw)   [WT](raw)   [KO](normalized)    [WT](normalized)

I have no idea where to start and woul dbe grateful for your help!

Thanks, Bhuvan

[Kevin Blighe]

Hey, so, you have a list of differentially expressed lncRNAs - how many? I suppose that the starting point is to correlate these to the other genes and to create a network from that correlation. Network analysis, while still popular to some people, has not proven to be of much clinical use, in my opinion.

So, you could also, for example, do the correlation between the lncRNAs and the other genes, take the statistically significant correlates, and then attempt to validate these by using the data on some lncRNA target database. I believe there are databases that have in silico and in vitro validated lncRNA-to-mRNA targets.

[lalithabhuvan]

Hi Kevin,

Thanks much for your reply. I have some 3000 lncRNAs that are differentially expressed. I will try to do what you said.

Thanks again, Bhuvan

[John]

Hi Kevin,

I have a count matrix from RSEM, triplicated samples for all three conditions. Sorry about lacking information. Can you check my updated question.

Thanks for your answer.

[Kevin Blighe]

I still believe you can achieve what you want via differential expression comparisons. You have not defined 'unregulated' or 'normal', but a comparison like this would help:

negative treatment / (positive treatment + control)

...and then filtering for the genes that are statistically significantly differentially expressed and have negative fold-changes.

The variance-based approach mentioned elsewhere in this thread is not entirely valid unless. However, if you wanted to identify genes that are 'constant' across your samples, then you would also include genes with low variance, but that could be a separate calculation.

Try to think it through.... there does not have to be a specific R package for everything.

[John]

What I did is:

Find DE genes between control vs positive treatment, control vs negative treatment. Then I subset the common genes in both, the I plotted heat map, I got what I wanted. It worked for now. Will try to do it in proper way.

Thanks for all the answers.