SNP Calling

Consed can detect SNPs (including indels) and calculate the SNP
quality (the probability that there is actually no SNP but all of the
non-reference bases are base-calling errors) using the method of Li
as implemented by MAQ (Li, H., J. Ruan, and R. Durbin, Mapping short
DNA sequencing reads and calling variants using mapping quality
scores. Genome Res, 2008. 18(11): p. 1851-8).

The list of SNPs can be displayed in consed.  The user can click on a SNP
to view the reads and bases at that location, or can repeatedly click
"next" to sequentially view locations of each SNP in the list.  Consed
can also generate in batch a VCF format report of SNPs. 

Our implementation differs from that in MAQ in that we always consider
the reference base to be one of the 2 possible alleles, regardless of
its observed frequency in the reads.  We assign prior probabilities of
1-R, 2R/3, and R/3 to homozygous reference, heterozygous, and
homozygous alternate respectively, where R=.001.  Our filters are
slightly different: No position is considered unless it has a
non-reference base (or indel) of quality 20.  We ignore any bases
with quality < 10, we ignore any reads with mapping quality < 10, and
we require at least 3 reads that pass these filters.  We
require at least one read to have a mapping quality 20.   We do not
discard snps flanking a potential indel nor do we discard clusters of
potential SNPs in a 10-base window since these filters were designed
to handle problems from an ungapped alignment which are not problems
with gapped aligners such as our phaster (P. Green unpublished). MAQ's
methods include a factor of 0.85**k where k is the number of
discrepant reads, but we do not allow this factor to get smaller than
0.85**20 regardless of the number of reads.