Notes for Monday Feb6th Meeting

[abaghela]

Hey people,

I will also try to put down what I remember of Amrits conversation with us.

• Look at differential network analysis. This can be used to see the differences in networks between the two conditions (asthma and normal).
• We can also see if we can try to group patients together based on methylation and rna-seq profiles. We can maybe create networks from groups of patients and use this differential network analysis again. This way we can see how the networks change among asthma endotypes (if any).
• Try to download the data using the GEOquery package. It will give you all the data really nicely. I can't seem to do it from whatever computer I am on.
• https://www.ncbi.nlm.nih.gov/pubmed/27473063

[abaghela]

So it looks like we want to do networky things...

This is kind of repetitive...

I am not actually sure how to create networks with differentially expressed CpG (DMCs). Seems like they use WGCNA to make their weighted correlation network to make modules. Then they use ingenuity pathway analysis to identify protein-protein interactions of proteins associated with each module. I want to know if we can find the genes associated with each CpG and directly create networks from there. I will research more about this CpG network idea.

Here are the approaches I am thinking. 1) Cluster patients based on their RNA-Seq and methylation profiles. Then, we can see which genes/DMC significantly drove the formations of those clusters. I know a couple tools that can do this. 2) Cluster the patients as above. Would we be able to create networks of the genes/DMCs and see how the networks vary between clustered patients. 3) Create networks with the differentially expressed genes AND the differentially expressed CpGs. We can compare the networks then.

Networks can be created with Network Analyst, which makes protein-protein interactions with genes. Clustering can be done with a whole bunch of tools.

[abaghela]

Notes from Meeting on Feb 6th

Emma

• adjusting for technical/confounding variables in the RNA-Seq Data
• Paper: did PCA before and after cleaning up.. saw that the PCA plots were quite different
• Paper: used linear models. If you know the confounders, then you can "regress" them out.
• MOTIVE: The paper says that the data has confounding variables
• We need a good method to address this.

QUESTION? Are there ways to clean up the methylation data?

[ppavlidis]

@abaghela About "I know a couple of tools...": Bear in mind that we want you to do your work in R, for the most part. Using additional (non-R) tools is fine if there's no good alternative (maybe ask first?), but it's then especially important to not treat them as "black boxes": understand how they work (assuming it's not something we teach in class). Thus I don't really recommend using NetworkAnalyst for analysis. If you just need PPI data, maybe it has that, but you can get it from a more primary source like BioGRID and work with in R. Beware that databases like STRING and BioGrid have a lot of types of "interactions" besides PPIs, so you need to filter accordingly.

I may have some comments about the network analysis parts, but I'll wait for you to flesh it out.