search

SchulzLab / STARE

TF analysis from epigenetic and Hi-C data
MIT License
16 stars 2 forks source link

Quantile normalization required? #2

Closed Andi964 closed 8 months ago

Andi964 commented 1 year ago

Hello and thanks for your improvements on the ABC model!

The "normal" abc model uses quantile normalized epigenomic data against K562 cells (and thereby defines the cutoff of 0.02). Does STARE also benefit from quantile normalized epigenomic input when using data from novel/different cell-types?

Best Andreas

DennisHeck commented 1 year ago

Hello and sorry for the late reply!

STARE currently doesn't provide any normalization on the input for epigenomic activity, it takes the provided values as they are. We don't have any expectations as to what metric is used for enhancer activity, may it be read counts, a fold-enrichment or p-value. In this context I want to note, that we observed differences between the CRISPRi-screens as to with which cutoff a recall of 70% can be achieved (that's how the 0.02 in the original publication was defined). So I'm not sure of how much use a normalization against the K562 data would be. The number of interactions that we got for different cell types with a cutoff of 0.02 was rather stable at least.

Best wishes, Dennis