Project Proposal for Team_Undecided

abaghela commented 7 years ago

67fc51710d2d7968cf30206a2a64e927fa99b29f

https://github.com/STAT540-UBC/team_Undecided/blob/master/project_proposal.md

@ppavlidis @singha53 @farnushfarhadi

singha53-zz commented 7 years ago

@abaghela no references?

singha53-zz commented 7 years ago

@STAT540-UBC/team-undecided

Some feedback:

Background:

I might be missing something here, but how do you know that the gene/methylation probes you identify from your differential network analysis are necessarily upstream of regulators of CLCA1, Periostin, and SerpinB2? Can they be downstream as well?
also explicitly state your hypothesis: Based on reading your background, it seems you are hypothesizing that molecular networks change between Th2 high, low and control subjects?

Division of labour

Seems fine, although since it is an assembly line type of set-up, it seems like the last person has to wait until the first 3 are finished. Is there a way to having different people working on different aspects of the project at the same time? Also it seems you will only use WGCNA but in your aims/methodologies, you state you will try many DiNA R-packages; be more specific here.

Dataset

I can tell which datasets you used from your background/Introduction, but it belongs in a different section (Dataset) just makes the marking easier ;) rubric

Detailed Aims/Methodologies

Aim 1: Normalization: It might be better to have 2 people work on the normalization (one for transcriptomics and one for methylation), because it seems like that is a rate limiting step. This might speed up the process.

Aim 2: Be as specific as possible when stating aims. It will make your life a lot easier when you do your analyses. For example, you state you will cluster subjects based on those 3 genes but which clustering algorithm will you use? How will be determine your boundary threshold?

Aim 3: Here you will apply WGCNA to each group separately resulting in three different networks. Will you be applying this on the entire methylation dataset (450K probes and 20K genes) or will you do some pre-filtering? Since you are clustering features (genes/methylation probes) it might take a long time if you keep everything.

Aim 4: I am unclear as to how you will determine if a network is differential; is that through module preservation statistics? If a module is not preserved across conditions does that mean it is not differentially expressed?

Which other DiNA packages will you try? I have used DINGO in the past because it identifies a network where the edges correspond to differential edges. That is, the partial correlations (a correlation between 2 nodes that is adjusted for correlation with other nodes; concept comes from Graphical Gaussian Models, using the glasso package under the hood) between all genes are calculated in each group. Then the partial correlations (in each group per node pair) are subtracted (signal) and divided by the standard error (noise). The only problem I found with DINGO is that it’s very slow! If you have more than 100 variables (e.g. genes) than it will take hours (because it needs to do bootstrap and re-estimate the differences to calculate the standard error). One way to mitigate this strategy is to only feed into differentially expressed genes/cpg probes based some stringent FDR correction. Also DINGO only works for a two way comparisons.
You can also hold the pathways constant, based on particular databases (e.g. KEGG pathways) and see how these changes across phenotypes.
explore the eigengene concept in WGCNA (summarises each module into a single variable)

Aim 5: pathway enrichment but you don’t mention how you would do this? Check out Bioconductor for potential gene set enrichment libraries. https://www.bioconductor.org/packages/release/BiocViews.html#___Software Also note that limma has gene set enrichment analysis capabilities!

Aim 6: is unclear. Be more clear about this. For example to which genes will you identify these drug targets for and why? You will identify many modules, that is a lot of drug candidates. I have used this tool in the past called Enrichr (http://amp.pharm.mssm.edu/Enrichr/). It is a web interface, but you can download the genesets (from the Libraries tab) and compute the enrichment statistics yourself using a fisher’s exact test for example.

Also you forgot this: Please provide a link to your project_proposal.md file in your group repo README.md.

Lastly, did your team decide on a name yet?

Looks great otherwise. Well Done! Happy to discuss further :)

santina / team_Undecided