I have been working on data analyzed by Samtools or GATK, and in both cases indels were badly detected. More especially, duplications or small insertions were missed by the two tools. I had already heard about this problem, but I had thought that I would rather see false positives than false negatives. Do you know any algorithm that performs better regarding this precise point?
Dear Anupam,
I thought that these solutions were dedicated to upstream annotations of variant files inferred from files parsed with a tool for variant calling, such as SamTools or GATK. Was I wrong?
Regards, Sébastien
Big question here, you are correct that indel are really tricky, even with Sanger sequencing of course. Certainly, we have difficulties with Samtools too. We get false positives certainly, but have not really assessed false negatives in a meaningful way.
So you know there is a particular indel in your case from other technologies? You message suggests this. Is your coverage of that region adequate?
There a reasonably useful review:
http://bib.oxfordjournals.org/content/early/2012/03/24/bib.bbs013.abstract
Good luck
Dear Sébastien,
What is the size of the variants you are interested in detecting?
If they are small indels, on the order of 1-25 bp, a good tool is Dindel (http://www.sanger.ac.uk/resources/software/dindel/). If you are interested in larger variants then the best tool(s) would depend on the data you have (i.e. coverage, read size, paired-end vs. single-end).
Margarida
Thank you all for your suggestions. It's very helpful.
Actually, I observed these false negatives in targeted exome sequencing. Read depth was good, i.e., above 100. This observation concerns indeed small indels about 10 bp, and more especially homozygous ones. To be more precise, the cited algorithms did not exactly miss everything; when looking at BAM files, I noted weird sequences not matching reference sequence at all and which looked like series of artifacts. These erroneous sequences differed when regarding forward or reverse strand. It looked all the more as artifacts as for a given homozygous duplication, these artifactual sequences barely represented half of the reads. What the algorithms detected was a series of 3-5 “false” variants with a poor quality call surrounding the true anomaly. I might post a picture later on to illustrate this.
Regards, Sébastien
Hi Alessandro,
I visualized BAM files with Alamut, but it works with IGV too. The difference just lies in annotation. The reference sequence is the one released in NCBI, Ensemble or UCSC.
Sébastien
Maybe this recent paper will help:
http://mbe.oxfordjournals.org/content/early/2012/12/04/molbev.mss256.full.pdf?papetoc
Hi Alessandro,
I think you're right regarding local realignment. I will start the analysis over, taking your remark into account to see if it performs better with your reference. Thanks, Sébastien.
Dear Dr. Sébastien Küry, Following are the accessible links for some articles. May give you any help.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2704429/
http://bioinformatics.oxfordjournals.org/content/early/2013/04/25/bioinformatics.btt198.full.pdf
- Badges
- Science method