fake indels from mpileup?

[MatthewVC]

Hey,

I'm running mpileup to generate a consensus sequence and when I run it and convert it to .vcf format, I see many indels that match the following sequence. My command line and an example of the odd indel is below:

samtools mpileup -Q 25 -d 40000 -uf ref.fa in.bam | bcftools view -cg - > out.vcf

gi|223976078|ref|NC_012095.1| 5988 . CTTTTTT . 99 . INDEL;DP=39134;AF1=0;CI95=1.5,0;DP4=14546,15887,67,67;MQ=60;PV4=0.66,1,1,0.028 PL 0
gi|223976078|ref|NC_012095.1| 5989 . T . 99 . DP=39199;AF1=0;CI95=1.5,0;DP4=18087,20153,0,0;MQ=60 PL 0
gi|223976078|ref|NC_012095.1| 5990 . T . 99 . DP=39116;AF1=0;CI95=1.5,0;DP4=17719,20059,0,0;MQ=60 PL 0
gi|223976078|ref|NC_012095.1| 5991 . T . 99 . DP=39395;AF1=0;CI95=1.5,0;DP4=16529,18557,1,0;MQ=60;PV4=0.47,2.4e-05,1,0.073 PL 0
gi|223976078|ref|NC_012095.1| 5992 . T . 99 . DP=39105;AF1=0;CI95=1.5,0;DP4=16713,18665,0,0;MQ=60 PL 0
gi|223976078|ref|NC_012095.1| 5993 . T . 99 . DP=38041;AF1=0;CI95=1.5,0;DP4=17015,19513,0,1;MQ=60;PV4=1,0.00047,1,1 PL 0
gi|223976078|ref|NC_012095.1| 5994 . T . 99 . DP=38046;AF1=0;CI95=1.5,0;DP4=17325,19918,2,0;MQ=60;PV4=0.22,5.4e-07,1,0.19 PL 0


As you can see, the "deletion" sequence "CCCCCC" matches the following sequence, making me think it's not real. I see this same pattern numerous times.

Any ideas what is going on here?

Thanks,
Matt

[swbarnes2]

I don't think the alt letter is supposed to be a .

What does the mpileup look like?

[eslondon]

In any case about 90% of the INDELs you will get form mpileup will result to be fake once you put them through tools like DINDEL and/or GATK.... don't trust those generated by mpileup, you need local re-alignment (and often visual inspection!) of the reads to get a high rate of true positives in terms of INDELs

[colindaven]

You'll also need to filter by the I16 fields in the output, once you generate them properly. I frequently use reference vs non reference quality sums. And yeah, every program I have tried - Samtools mpileup, dindel, Clcgenomics - seems to produce a lot of false positive indel calls.

[MatthewVC]

Thanks for the replies.

I piped the mpileup directly into bcftools, so I don't have the mpileup file handy. I was curious about the "." too. I think it's ok, since where there are different calls, the "." is the alternative call.

I'll try filtering the file through the DINDEL or GATK. I may just visually go through, and then write a script so sort the fake indels out. Looking at the .bam files in IGV, I'm not seeing any of these indels, so I really think they're pretty much all fake.

Thanks,

Matt