search

ZikunY / CARMA

GWAS genetics Fine-mapping method
GNU General Public License v3.0
20 stars 7 forks source link

Error in C.list[[2]][[ncol(set.gamma[[i]])]] #22

Open mconery opened 6 months ago

mconery commented 6 months ago

Hi,

We're excited to use your CARMA tool for a new analysis that we're running, but during our initial testing, I came across the following error with one of the loci we were mapping:

Error in C.list[[2]][[ncol(set.gamma[[i]])]] : attempt to select less than one element in integerOneIndex

Do you have any idea what's causing this error? I ran this locus alongside a bunch of others that finished without issue, so the format of the inputs shouldn't be a problem. I think it's something about the z-scores and LD correlations that's causing the calculation to break down. In case it matters, I ran the algorithm without annotations, with the outlier check on, and given a rho value of 0.95.

If I can provide any detail, please let me know! Thank you in advance for the help!

Mitch

ZikunY commented 6 months ago

Hi Mitch,

Thank you for choosing CARMA!

The error is related to the part of the candidate models that have been visited by the Shotgun algorithm. However, without actual data, I can not tell where is the problem. Did CARMA finish this locus properly without the annotations? If so, maybe reduce the number of annotations to check if there was any annotations causing the problem.

How many annotations were provided to CARMA?

Best, Zikun

mconery commented 6 months ago

Hi Zikun,

Thank you for getting back to me so quickly, and I apologize if I was unclear! I ran CARMA without annotations the first time, and it errored out.

Interestingly, I just tried rerunning CARMA with a different random seed but otherwise the same setup, and it worked this time. This outcome leads me to believe that, like you suggested, the issue is probably something very specific to how the input data is interacting with the algorithm. If you want to dive into the issue further, I could try to find a way to get you the data.

Thank you, Mitch

ZikunY commented 6 months ago

You are welcome! I guess the problem could be the outlier detection dropped quite a lot candidate models at the very beginning of the algorithm, leading to empty library of the visited candidate models, which caused the problem.

It would be much appreciated that if you could send this particular example to me, my email is yangzikun1125@gmail.com. I think as long as the errors are not consistently and repeatedly shown up, I probably not gonna fix it immediately but will keep it in my mind.

Thank you for letting me know.

Best, Zikun