Closed xiaohk closed 1 year ago
If we find the same results - that the images do not separate by compound in the t-SNE plot - we could also construct an experiment where the images should look different from one another. That might involve identifying compounds in advance that are known to have a strong effect on the cells using prior knowledge about the mechanism of action and then visualizing the network-transformed images for those compounds instead of the set of compounds on a plate.
aedebfeeef977e41038bde2732009e9c688d9746 adds result of all t-sne plots of 16 plates.
There are some interesting plots:
I visualized j16
in plate 24280
. It looks there are few cells in the image: