Also not sure if we ever got an answer to this but it would be very important to know whether cells from the same sample could have ended up in both the “training” and “test” set for the CV—if so, it’s violated the core assumption of independence and none of the accuracy measures can be taken at face value.
This quote from Dr. Xu in the webinar seems to say that they did not allow cells to be split up:
Our approach is definitely patient level split. So the single cell[s?] from one patient will just belong to that patient and will not enter both the training and test sets and also we split the patient into train and test so that one patient never enter both of the sets.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
