Closed graceleuchtenberger closed 4 months ago
Regarding where to go from here, @sr320's comment still applies:
Thanks Sam! I think Matt wanted to see a PCA across all four treatments first, so that's why I did what I did. Hoping to get a go ahead to see if a pairwise comparison is worth it given the graph above.
Yes - I say pairwise info is worth getting.
Nice work Grace!
As far as the PCA goes, some improvement could be:
Agree that next steps are trt vs control comparisons of gene expression. The PC1 vs PC2 axis didn't pick up much of the variance, so there is likely a lot of overlap. A list of genes that are differentially expressed across each treatment relative to the true control. I would start with just significant and not filter by lfc to start.
Here's the new graph with the true control. This is just to update you all, I'll keep on with the pairwise comparisons. "Control_3" is day 3 of the control treatment. The pairwise comparisons for the PCA were:
[1] "Intercept" "treatment_control_3_vs_control"
[3] "treatment_DO_vs_control" "treatment_OA_vs_control"
[5] "treatment_OW_vs_control"
Wow, really interesting! Looks like bringing them into the lab really made a difference.
I did pairwise comparisons (true control to everything else), got PCA's and DEG lists. See the PCA's here
Amazing work @graceleuchtenberger!
Agreed with @mattgeorgephd - very nice!
Some interesting stuff in those PCAs (e.g. foot shows clear differences in response to OA, but gill do not...)
Hi Grace! Jumping in here to say cool plots and results! I have a question for my own knowledge - would it make more sense to do a 2 step analysis: 1) DEGs/comparison between control 0 and control 3 to account for tank effects followed by 2) control 3 to all other treatments. If you do a comparison between control 0 and all other treatments, your differential effects are going to be confounded by the change they experienced being brought into the lab. Is a more relevant comparison actually control 3 to all other treatments on that same day?
Agreed, a two step analysis would be helpful here. Here is the approach we came up with. Interested in everyone's input:
I think this accomplishes the same thing as the approach that @AHuffmyer suggested, but has the potential to more accurately capture tank level effects as they relate to animals in each treatment. Treatments in this case were not pseudo-replicated by splitting animals into different chambers, so we really are looking at tank level effects.
I finally got a PCA to work with some code from Sam's snow crab repository. Here's the plot:![Screenshot 2024-02-05 at 2 09 59 PM](https://github.com/RobertsLab/resources/assets/122406581/fb5e4c8d-5d13-4b3f-a98c-71d3d836ca5a)
See the overall code here.
Here's the code section I used to generate the plot:
Visualize differences between treatment groups
Where do people want to go from here? Did I do the PCA correctly?