Normalizing imputed spatial dataset

The inputed spatial transcriptomics data is only available in log2 transformed form. Some downstream analysis requires normalized data. What is the best way of doing this? Can I get back to the raw counts by 2^(log2 values)?

Hi @wykpenguin ,

According to the scientists who did this analysis from the original publication:

“We used the log normalized value (y) in the imputation, which is log2(CPM + 1), where CPM = (10^6 /sample_count_sum) * rawcount. Imputation takes a knn average of these values. So the exact reverse engineering of rawcount = ( sample_count_sum / 10^6) \*(2^y - 1) won’t be feasible. Since we are taking knn averaging, using rawcounts might not be desirable, [and] we don’t have the rawcount for Merfish data.”

If you need additional information, please reply to this thread.