search

guanjue / S3norm

An normalization method that can normalize high signals without inflate the background noise.
MIT License
7 stars 3 forks source link

Activating AND repressing modifications in same dataset #4

Open rlh60 opened 2 years ago

rlh60 commented 2 years ago

Hi, please can you tell me if S3norm is suitable to be applied to many samples containing very different histone/chromatin modifications (including both activating and repressing, e.g. H3AcK27 and H3Me3K27) and very different cell types?

We don’t get nice correlation heatmaps and have concerns that there may not be many common peaks between our samples and if there are any, due to the very different nature of the histone modifications used in the analysis, those peaks won’t be the most appropriate ones to use to normalise the samples.

We would very much appreciate your thoughts about this and would be extremely grateful if you had any ideas of how we could improve our results.

Thanks in advance! Rebecca

guanjue commented 2 years ago

Hi, Rebecca

If there is just small number of common peaks, it is probably not fit the assumption of S3norm, so the normalized signal may not be good.

Depends on what is your downstream analysis, you may need different normalization strategy. But based on my experience, the quantile normalization using the average signal as the reference is usually the most robust normalization strategy. (The quantile norm will force different sample to have the same distribution which will create false positive/negative signals.)

Best wishes. Guanjue