Open niranjchandrasekaran opened 2 years ago
Platemap and metadata files are in the metadata folder.
I calculated the replicate correlations and Percent Replicating for each cell type separately.
U2OS has the lowest scores while the other three have much higher scores. Each dose point is considered a separate sample
Description | Percent_Replicating |
---|---|
Cell Type=PATU-8988T | 51.5 |
Cell Type=MDA-MB-231 | 57.6 |
Cell Type=U2OS | 39.4 |
Cell Type=A549 | 66.7 |
I plotted replicate correlations as bar plots.
In order to be able to interpret whether a replicate correlation value is significantly different from the null distribution, here are the null thresholds
Description | Value_95 |
---|---|
Cell Type=PATU-8988T | 0.383839 |
Cell Type=MDA-MB-231 | 0.305252 |
Cell Type=U2OS | 0.177008 |
Cell Type=A549 | 0.285751 |
The following are the MOA of the compounds which will perhaps help interpret the results
Supplier_Catalog | moa |
---|---|
FMF-03-146-1 | DCLK1 inhibitor |
FMF-04-112-1 | DCLK1 negative control |
FMF-04-159-2 | TAIRE kinase inhibitor covalent |
FMF-05-176-1 | TAIRE kinase inhibitor reversible control |
BSJ-04-146 | undisclosed target PROTAC |
BSJ-05-002 | undisclosed target PROTAC negative control |
BSJ-04-175 | undisclosed target parental inhibitor |
Pin1-3 | Pin1 inhibitor covalent |
JWG-071 | ERK5 inhibitor |
AX15836 | ERK5 inhibitor |
JWG-119 | ERK5 inhibitor negative control |
Replicate correlation seems to increase with concentration, which is good. A couple of other observations I made are as follows
As I mentioned above, these compounds don't seem to be very active in U2OS.
I stacked the bars so that overall effect of the compounds across all cell concentrations and cell types can be visualized.
I used the replicate collapsed gct file to create this morpheus save session. I grouped the heatmap by cell type. There are number of interesting blue areas all over the heatmap. In general, they seem to correspond to the BSJ or FMF compounds.
cc @AnneCarpenter
I added all the controls and the updated gct file and morpheus session are here and here respectively.
I tried using euclidean distance as the similarity metric for clustering. I clipped the values at 200 so that all the shades of red are visible. I am not sure if this heatmap is more informative than the one above.
Thanks! I agree that adding DMSO to these plots, when it's been normalized almost to oblivion and has a very low replicate correlation is not very helpful because it could easily land in any of the clusters, just by slightly changes in the profile relative to 0.
Goals: Exploratory analysis to ensure everything is working as expected and look for relationships between perturbations
Conclusion: