Hi Pablo,
Say one uses dge raw data to fit the x,y,z coordinates of the 1297 cells using some of the 9000 genes available, and does not overfit.
Q1) Is is fine to list in the submission file the most important say 20 of the insitu.gene names that may be among predictors, yet we understand that they are not the only predictors used to predict the x,y,z location of the cells (since no insitu matrix data is used)?
Q2) Is it ok to look through the original geometry matrix (3039x3) to selected the closest 10 locations of the predicted x,y,z values for a given cell? Or is one limited to use only the geometry matrix for the 1297 cells, when picking the 10 closest neighbors .
Q3) since no insitu matrix data is used in the prediction, one can submit the same predictions for all three sub-challenges but list increasing sets of the insitu.gene names. Is this ok?
Thanks,
Adi
Created by Adi Tarca bcbuprb Can you please clarify what an adequate rationale is? I would assume that the top-X closest is indeed a rationale and is based on a predictive model. Thanks Pablo Hi Adi, Q1 and Q3 seem fine to me of course in Q2 you need to have a rationale for how you chose the 10 cells, not only the 10 closest physically to your top prediction
P