False negative variant calling in haploids. Call variants using coverage (not stats)?

[Genomics101]

Greetings,

I am looking for the simplest possible way to call variants computationally, just using position coverage and the ratio of reference/alternative allele, for divergent haploid genomes. I.e. no need for prior probability or a prediction based on an allele frequency spectrum. There is good coverage (average depth of coverage is 316X), and the sequences are 100bp reads of high-quality.

I am mapping with BWA and using bcftools to call variants using Bayesian inference in "mpileup" files. It seems bcftools is just too fancy for this to work among these divergent haploid genomes, and I am getting only half the variants I should be, even setting the quality scores as low as 5 for filtering.

Looking at my BAMs, the variants are there but are getting removed somehow, probably because they don't fit with some expected allele-frequency spectrum due to the level of divergence (~10-20 variants/kb)

Is there a simple way, using bcftools, a script that goes into the pileup, or another program to get better calls? FreeBayes is recommended here for haploid genomes, and I'll look into it, but I think it may end-up with the same problem.

Thanks!

[adaptivegenome]

I am assuming by variants you mean SNPs. FreeBayes is good. Also try JGIL.

[Genomics101]

SNPs and small indels, yes. I'm looking into FreeBayes now. I also ran the analysis on each BAM individually, rather than using a population sample, with bcftools, and the results were much better.

[BAMseek]

Here is another seqanswers thread that might give some ideas if you haven't already seen it:

http://seqanswers.com/forums/showthread.php?t=10182

If your looking for a simple way to call SNPs from the base frequencies, you might check out

http://protocols.netlab.uky.edu/~liu...rs/Vijay_1.pdf

where they call something a homozygous SNP if the non-reference allele is >80% (after filtering on quality Q20 and coverage 20X). They called hets if the alternate allele was between 20-80% (which wouldn't be pertinent for you since you are dealing with haploids). This could at least serve as a base line sanity check.

The IGV browser employs a similar heuristic to show variants:

If a nucleotide differs from the reference sequence in greater than 20% of quality weighted reads, IGV colors the bar in proportion to the read count of each base

Justin