Questions about survival prediction and K-means initialization

[dashuaige338]

Hi,

Thanks for your great work! I have two questions here for you. I was trying to run a survival prediction using PANTHER + LinerEMB (Step2B) on my own dataset based on extracted UNI features. However, the outputs seemed to make no sense to me, as I got an empty summary.csv and an all_dumps.h5 without signature. I also cannot find a detailed guide for survival prediction workflow, so could you please provide a workflow for survival prediction and specify what would be the expected outputs?

Second, for your K-means initialization, I saw you didn't specify a random state here, which means multiple times of initialization on the same dataset might return different prototypes. In this case, I think it would be interesting to compare the similarity of prototypes generated by myself and those in your paper at patch or WSI level. So would you mind upload your original prototype .pkl files for TCGA LUAD & LUSC somewhere? I'm also interested to see if the labels for your 16 prototypes in the paper work for my prototypes.

Thanks!

[andrewsong90]

Thank your for patience - I am currently out with some personal tasks, but will get back to this shortly.