In the provided notebooks, the scale factor in log1p normalization in Crunch1 for Xenium data is 100 while the scale factor is 10000 in Crunch 2 for scRNA-seq data. Can someone confirm this? The task for Crunch 2 is to predict gene expression in held-out Xenium data, but why would we need to use a scale factor of 10000 in the submission?
Yes, we confirm that the scale factor is 10,000.
In Crunch 1, you had to predict 460 genes; in Crunch 2, you have to predict 2,000 genes. Initially, there were 18,615 genes, but for performance reasons, we reduced it to 2,000 genes. However, in the first version of the paper(full specification), the scale factor was 10,000, so we kept it.