Rverse perturbation results
Hi !
I have another question concerning insilico. Sometimes the number of start state cells available for forward perturbation is limited. I wonder whether it is valid to instead use the reverse perturbation results from goal state cells (specifically the ones with cosine similarity < 0) as an equivalent proxy?
Thanks for your question. It is likely that many perturbations would exhibit the opposite effects, though genes can be context-specific so it may not always be valid to simply reverse the expected effect. Our experimental validation has focused on the perturbation set up in the forward direction, but it makes sense that one may wish to run it in reverse if the number of cells is limiting, but the average goal embedding position is still certain enough.