search

ay-lab / FitHiChIP

Statistically Significant loops from HiChIP data
MIT License
39 stars 20 forks source link

Lower number of Significant Interactions #98

Open koushik20 opened 1 year ago

koushik20 commented 1 year ago

Hi, First thank you for providing excellent documentation on FitHiChIP!

I am running FitHiChIP on my HiChIP data generated from HiCPro (.allValidPairs). I am using the following parameters.

IntType = 3
BINSIZE = 5000
LowDistThr = 5000
UppDistThr = 1000000
UseP2PBackgrnd=0
BiasType=1
MergeInt=1
QVALUE=0.01

I am using several Breast Cancer HiChIP data with H3k27ac peak files (No.of peaks 30k - 50k) while the Interactions considered ranged from 400k to 700k the significant interactions found were in the range of 300 - 650. Why am I able to get so less of significant interactions? Is there any other way to find the reason for a smaller number of significant interactions being found?

Thank you!

ay-lab commented 1 year ago

Hi @koushik20 Thanks for your query. I've added one script QC.r under the folder "QC" in the FitHiChIP repository. Please execute this script and check the "out.log" file and the associated pdf. My understanding is that the average contact count for the interactions in your data is quite low, resulting a lower number of significant interactions. For example, in one of our good quality HiChIP data, P2P=0, coverage bias, distance = 10 Kb - 3 Mb complete number of contacts (interacting pairs) = 1980378 number of significant contacts = 62490 average CC for the complete set of contacts = 4.5, average CC for the significant contacts = 25.9 Also check the file "Distance_vs_AvgCC_Complete_Contacts.pdf" which shows the average CC for the complete set of loci (with nonzero chromatin contacts) subject to individual genomic distance binning. In this dataset, the average contact never drops below 3 even for 3 Mb distance.

Please run this script on your data and check the output log file and plots, to see if the sequencing depth and / or the contact count distribution is an issue in your data.