Problem with guess in emStep

Question

Problem with guess in emStep

tristanpwdennis opened this issue 5 years ago · comments

Hello
I am trying to bootstrap my relatedness estimations.
I am using the command: ~/software/NgsRelate/ngsRelate -f freq -g angsdput.glf.gz -O test -L 521328 -n 81 -a ind1 -b ind2 -B 100

For most of the files (I am looping over ~200 individual pair combinations which are a subset of the whole callset contained in the .glf and .maf files) I get this error:

Problem with guess in emStep: -nan -nan -nan -nan -nan -nan -nan -nan -nan

I get the error when I loop over a list, and also when I specify individuals as a single command.

For 5 of the comparisons it seems to work fine (estimation has been made 101 times, as I guess the bootstrapping is zero indexed...). Can't work out what is going on.

Please could I trouble you for some assistance?

I have attached a Drive link to a compressed folder with the data and the analysis script (bash wrapper):
https://drive.google.com/open?id=1eraeCDVdOa-idliPg00TgMDkeRLM8zTz

Thank you!

Tristan

KHanghoj · Answer 1 · Wed Dec 04 2019 01:45:48 GMT+0800 (China Standard Time)

Hi Tristan,

Sorry for the late reply.

Thanks for providing an example upfront. it made it a lot easier to fix the bug. The bug was related to sample with replacement error.

When you provide -f freq, you dont have to also set the number of sites -L. Furthermore the example had 318194 sites and not 521328.

Let us know if the update also works for use.

Kristian

Tristan Dennis · Answer 2 · Wed Dec 04 2019 19:31:59 GMT+0800 (China Standard Time)

Hi Kristian
Thanks for getting back.
That seems to be working really well now, thank you.

I have one final question:

I notice in the output of the bootstrapping from the example data, the rab and KING values (that I have been using as the relatedness estimate) is now between 0.99 and 1 - even when the two individuals are, say, half siblings (so I expect ~0.25 or so...). However, a number of the other coefficients in the output file (J1, inbred_relatedness, fraternity) are more reflective of the true relatedness (some of the samples I know the relationship already). I've added the results file below.

This only happens when I use the bootstrapping - ought this to be happening?

Thank you for your assistance again!

Tristan

C1_S25-C4_S28.res.txt

KHanghoj · Answer 3 · Tue Dec 10 2019 23:36:36 GMT+0800 (China Standard Time)

I have just tested ngsrelate.

i get rab of 0.18.

When you only analyze a single pair using -a and -b, you need to provide the index (0-based) of the two individuals.

./ngsRelate -f freq -g angsdput.glf.gz -n 81 -a 24 -b 27 -O test2_boot  -p4 -z list -B10

If you provide sample names to -a and -b, the software will take the first individual twice :)

Let us know if it helped.

I'll add a check for this in ngsrelate

Kristian

KHanghoj · Answer 4 · Tue Dec 10 2019 23:59:22 GMT+0800 (China Standard Time)

I have now added a sanity check. if a string is provided as argument to -a or -b, the software will return an error.

Let me know if it works.

Tristan Dennis · Answer 5 · Tue Dec 17 2019 22:02:39 GMT+0800 (China Standard Time)

That works fine, thank you!

However, I pulled the most recent and passed 2 strings for names instead of indices and no error message
~/Software/NgsRelate/ngsRelate -f freq -g angsdput.glf.gz -n 81 -a C4_S28 -b C1_S25 -O test2_boot -p4 -z list -B10

KHanghoj · Answer 6 · Wed Dec 18 2019 01:09:26 GMT+0800 (China Standard Time)

Good.

I have just tried your command:

../NgsRelate/ngsRelate -f freq -g angsdput.glf.gz -n 81 -a C4_S28 -b C1_S25 -O test2 -p1 -z list -B 1
and get the following error:
pair 'C4_S28' is not a 0-based index

Ill close the issue. feel free to reopen if you encounter other issues with ngsrelate.

Glad that you're using the software.

Kristian