search

rr1859 / R.4Cker

MIT License
16 stars 15 forks source link

Error when running Cis analysis function #8

Open tlgolan opened 8 years ago

tlgolan commented 8 years ago

Hi, When I am running the mentioned function, I get this error:

cis_results=cisAnalysis(my_obj,k=10) Error in .local(object, ...) : Initial model infeasible, log likelihood is 'NaN'; please provide better starting values.

Can you think about a reason?

Thanks

rr1859 commented 8 years ago

Have you looked at your raw data? Do you see a decrease in signal away from the bait?

tlgolan commented 8 years ago

I do see a decrease in signal as a function of distance from the bait. Do this package assume anything about the mapping quality of the reads and the length of fragment ends? Can you think about a reason why I can complete the near-bait analysis and not the far-cis one? Thanks for your help!

rr1859 commented 8 years ago

No it does not - but have you checked if your sample passes the QC criteria? (as per van der Werken 2012). Also have you tried different values of k?

tlgolan commented 8 years ago

I read that the recommended values of k are 3-5 for near cis-analysis and 10 for far-cis. Do you have any other recommendations? Thanks!

tlgolan commented 8 years ago

Repeating my question here:

I read that the recommended values of k are 3-5 for near cis-analysis and 10 for far-cis. Do you have any other recommendations?

Thanks!

rr1859 commented 8 years ago

If you look at supplementary figure 2 in the paper you can see the differences between several values of k for the cis analysis. The value of k determines the number of observed fragments in each window, so it really depends on your the coverage of your data. By increasing or decreasing the value you will either get interactions or high or low resolution.

Also as I asked above - did you check if your samples pass QC criteria?