Region score question about cistopic

[cbravo93]

Hi @thouis !

There is indeed an error in the formula in the paper and it should be, as written in the code:

Thanks a lot for noticing this error, we will contact the editor to correct it in the paper.

The region scores indicate the importance of each region to each topic (each region will have a score in each topic) and are used in different steps. They are calculated with getRegionsScores() and stored in [email protected]. We use them directly when using AUCell for checking the enrichment of epigenomic signatures (regions are ranked based on this score, in signaturesHeatmap()), and for the dimensionality reduction (where by default we only use the binarised regions). For the rest of the steps (e.g. GREAT, cisTarget) we binarise the topics taking the top regions based on these scores (with binarizecisTopics()), and use this 'top set'. I will go over the documentation to try to make this clearer.

Hope this helps, let me know if there are more questions!

C

[thouis]

Thanks for the clarification. Is this metric from the LDA literature? I have a general sense how it works, but not a very good one, yet.

[cbravo93]

Hi,

This metric specifically is also used in the R package lda for ordering words within topics: https://github.com/slycoder/R-lda/blob/master/R/top.topic.words.R .

Cheers,

C