Method to detect gene co-expression of genomic feature in cell lineage

Question

[gundalav]

I have data from the cell lineage as samples:

Cell1 -> Cell2 -> Cell3

Then for each of the above samples I obtained the expression for 'coding' and 'non-coding' genes.

What I want to do is to identify cluster of genes for each cell lineage above, where the cluster should satisfy the following category:

Cluster Type 1:
non-coding UP
coding UP

Cluster Type 2:
non-coding DOWN
coding DOWN

Cluster Type 3:
non-coding DOWN
coding UP

Cluster Type 4:
non-coding UP
coding DOWN

What I mean by UP and DOWN is the expression trend through cell lineage.

Is there any method to perform such kind of clustering?

Answer 1

[Irsan]

Assuming that you have adequately processed (including normalization !!!) your raw data to gene expression estimates, so

• for micro-array used e.g. RMA-normalization and you are using log-transformed normalized values
• for RNA-sequencing you used negative binomial approach (edgeR, DEseq2, ...) to get log-transformed CPM/RPM/FPM (or whatever you want to call it)
• By up, you mean that gene expression linearly increases: cell1 < cell2 < cell3
• You have used the proper and dedicated statistical methods/design matrix (linear-by-linear / time-course) to identify features that significantly linearly increase with these 3 conditions
• You understand that 3 conditions (by far!) is not enough to asses linearity

Then I would suggest that you use hierarchical clustering of both features and samples of scaled log-transformed expression estimates of features that significantly change between your conditions. From the resulting heatmap, you can judge whether clusters of genes with your preferred expression profile are present in your data set.

If you are not sure whether all the assumptions hold (or if you don't understand), I would strongly suggest that you fix that first. If you are not able to do so or you just want a quick solution, find a (local) collaborator who can.