Suggestion for modelling transcriptomics analysis
0
0
Entering edit mode
7 days ago
greyman ▴ 120

I got two sets of transcriptomics data (1day & 2 days exposure time) which consist of 20 samples each, with 4 different chemical concentration each and 5 replicates per concentration. Overally there are 40 samples. I used linear regression in deseq2 to examine the effect of different concentration (numerically) and basically found Degs for each data set. There is a batch effect due to different exposure time therefore i did not carry out time series analysis. There are comments that gene expressions which dont follow a linear pattern will be missed out from this analysis and this analysis is very uninteresting (boring). It makes me wonder if i should check out for tools that can do different types of modelling(any suggestion on tools? I use R ) or try to do machine learning like WGCNA to find diferent modules? More suggestion /opinions / criticisms are really appreciated.

rnaseq wgcna linearmodel deseq2 • 107 views
0
Entering edit mode

This ends up being the same as a time-series experiment, so have a look at http://bioconductor.org/packages/devel/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#time-series-experiments In general though people tend to cluster their data and have a look for obvious patterns in expression, which can then be used to make groups of genes.

0
Entering edit mode

Thanks for the reply. I did try with exploring time-specific differences of gene across all different concentration a few weeks ago, with the following code:

dds <- DESeqDataSetFromMatrix(data, metadata, design = ~Days + concentr + Days:concentr)

dds <- DESeq(dds, test = "LRT", reduced = ~Days + concentr)


Then i realised the data were bound to the batch effect, which is convolved with the effect of time (the 1st day and the 2nd day data were sequenced in different batch). It produced some really weird results, called a only a few DEGs, and the heatmap profile doesn't make any sense ( if you are interested to look at them I can put them up here). Also, I let "concentration" stay as a numeric variable...