DELLY Filtering Question for Somatic Variants
805 views
Skip to first unread message
Matthew Field
Dec 1, 2016, 3:15:37 PM12/1/16
to delly-users
I have gone through the tutorial for Delly on the Github page for filtering SVs, and I had a few questions.
1. When taking one sample and filtering it with a matched germline with tumor, to look for deletions and translocations I get 15,607 deletions and 128,326 tanslocations. Are these variants that are included in both germline and tumor, just tumor, just germline, or all three?
2. Because when I do the pre-filtering step with the text file that labels the samples as control or tumor, the number of variants go from 15,607 to 127 deletions and from 128,326 to 943 translocations. This seems like a huge amount of samples that are pre-filtered out, many with pass and precise as labels.
3. Then, when I do the re-genotype and post-filter, the 127 goes to 21 deletions and the 943 goes to 200 translocations. However, most of these locations that go through all the filtering are labeled as Low_Qual and/or Imprecise. I know that you have described before that these are related to mapping quality and number of reads, but do you believe the Low_Qual and Imprecise hits if they make it through all the filtering steps to this point? I ask because some of the remaining hits are very interesting. Do these numbers seem realistic?
4. I am also getting 0 insertions meeting the pre- and post-filtering for all my samples. Does that seem real?
5. Also, I used the exclusion regions provided with delly, but many of the remaining hits are very close to the centromere. Do you believe them if they are close to the exclusion regions but not in them?
6. Is there a good way with delly-suave to visualize translocations on both chromosomes, as opposed to just the originating chromosome? For example, it is hard to browse through the chromosomes and see translocations on chromosome 1 when they do not appear there visually and only appear on the other chromosome.
Thanks for your help,
Matt
1. When taking one sample and filtering it with a matched germline with tumor, to look for deletions and translocations I get 15,607 deletions and 128,326 tanslocations. Are these variants that are included in both germline and tumor, just tumor, just germline, or all three?
2. Because when I do the pre-filtering step with the text file that labels the samples as control or tumor, the number of variants go from 15,607 to 127 deletions and from 128,326 to 943 translocations. This seems like a huge amount of samples that are pre-filtered out, many with pass and precise as labels.
3. Then, when I do the re-genotype and post-filter, the 127 goes to 21 deletions and the 943 goes to 200 translocations. However, most of these locations that go through all the filtering are labeled as Low_Qual and/or Imprecise. I know that you have described before that these are related to mapping quality and number of reads, but do you believe the Low_Qual and Imprecise hits if they make it through all the filtering steps to this point? I ask because some of the remaining hits are very interesting. Do these numbers seem realistic?
4. I am also getting 0 insertions meeting the pre- and post-filtering for all my samples. Does that seem real?
5. Also, I used the exclusion regions provided with delly, but many of the remaining hits are very close to the centromere. Do you believe them if they are close to the exclusion regions but not in them?
6. Is there a good way with delly-suave to visualize translocations on both chromosomes, as opposed to just the originating chromosome? For example, it is hard to browse through the chromosomes and see translocations on chromosome 1 when they do not appear there visually and only appear on the other chromosome.
Thanks for your help,
Matt
Tobias Rausch
Dec 2, 2016, 7:11:55 AM12/2/16
to Matthew Field, delly-users
(1) delly call ... doesn't know about tumor or germline so it simply discovers abnormally mapping paired-ends and for small InDels, clusters of soft-clips. So the initial SV files indeed contain in your case somatic SVs, germline SVs and false positive SVs that are due to repeats and mis-mappings. For translocations, you might want to use delly call -q 20 ... which gets rid of most repeat-induced SV calls.
(2) delly filter -f somatic ... is of course tumor and germline aware and thus results in a much shorter list of somatic SVs. The expected number of somatic SVs totally depends of course on the tumor and the tumor re-arrangement patterns but having only a handful of somatic SVs is not unusual. In fact, I am pretty sure that among the 943 candidate somatic translocation calls you still have a large number of false positives, which is one of the main reasons why I recommend filtering somatic SVs using a panel of control samples as outlined in the README (https://github.com/dellytools/delly)
(3) The numbers are certainly realistic, what is more disturbing is that most remaining SVs are labelled as LowQual. That's indeed unusual and would only make sense to me here if you have a tumor of very low tumor purity or very high heterogeneity or some other technical issue in the sequencing data.
(4) Insertion sizes are bounded by the read-length in Delly and Delly only focuses on the larger insertions (>15bp). Thus, it's a small size band and the sensitivity is not that great because paired-end mapping doesn't work here. I am not too surprised if you get nothing here for somatic insertions. This probably needs some improvement on my end.
(5) That indeed makes the SV calls questionable. If you want to be more conservative here you can check if the SV calls fall into a GIAB callable region or not.
(6) Thanks for catching this. This should be fixed, we will work on that.
Best, Tobias
--
You received this message because you are subscribed to the Google Groups "delly-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to delly-users+unsubscribe@googlegroups.com.
To post to this group, send email to delly...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Reply all
Reply to author
Forward
0 new messages