search

3DGenomes / binless

Resolution-independent normalization of Hi-C data
GNU Lesser General Public License v3.0
7 stars 2 forks source link

Interaction and difference detection #13

Closed olechnwin closed 5 years ago

olechnwin commented 5 years ago

Hi, I have a question regarding the method to detect interaction and difference. Is there p-values associated with the significant interactions and differences? I cannot find this information from the paper. Or are you suppose to filter interactions based only on the fold-change? Thanks!

yannickspill commented 5 years ago

Hi, thanks for your question. Have you checked issue #3 ?

olechnwin commented 5 years ago

@yannickspill, yes I did. In that issue, you mentioned that interactions are significantly different from background if they are nonzero. My question is there p-values associated with the significant interactions?

yannickspill commented 5 years ago

@olechnwin This method is based on Bayesian inference (as opposed to frequentist statistics) and uses shrinkage as a means to control false positive rate. There is no p-value associated to each value of the signal/difference. If you seek to play with a theshold, you are allowed to increase the values of lambda2 and lambda1 in signal/difference detection. That will make the matrices more smooth (resp. keep data with larger fold changes only) and therefore reduces the false positive rate.

olechnwin commented 5 years ago

@yannickspill, thank you for your explanation.