Does low replicate correlation necessarily mean no signature?

[MarziehHaghighi]

Does low replicate correlation necessarily mean no signature?

[MarziehHaghighi]

Source Notebook

I start with a toy example:

• Here we have four gaussian blobs imagined to be all the profiles in a plate, each point corresponds to a well level profile. Points are 4d but visualized in 2d. The blob centered at (0,0,0,0) corresponds to control wells and has unit variance in all dimensions. Since we are z scoring all the wells to control wells. The other blobs are populations with a signature wrt control wells.

• 

• Now, lets calculate pairwise replicate correlations for all wells (most left subplot) and also wells inside each of the blobs.

• 

• Red vertical dotted line corresponds to median of the distributions.

• Class 0 corresponds to the control wells. Median of class 0 is higher than class one and lower than classes 2 and 3. There is no pattern in that regard. Standard deviation of all blobs are equal which means there is no difference in the degree of difference between samples of control wells and the wells in different clusters. Based on this figure, it is not true that if we are in class 0 the differences in the points result in a huge difference in the pairwise correlations comparing to the points inside the rest of the classes.