I don't completely understand. You have a partial reference, and you have SNPs from mapping to that, but you also have a load
of stuff that doesn't map and you want to call markers in that?
The Cortex assembler can do this. In fact you might be interested in this related paper, just out. It's not directly applicable as they sequenced paired-end, but
might be useful.
Reference-free SNP discovery for the Eurasian beaver from restriction site–associated DNA paired-end data
Helen Senn1,*, Rob Ogden1, Timothee Cezard2, Karim Gharbi2, Zamin Iqbal3, Eric Johnson4, Nick Kamps-Hughes4, Frank Rosell5, Ross McEwing1
In this study, we used restriction site–associated DNA (RAD) sequencing to discover SNP markers suitable for population genetic and parentage analysis with the aim of using them for monitoring the reintroduction of the Eurasian beaver (Castor fibre) to Scotland. In the absence of a reference genome for beaver, we built contigs and discovered SNPs within them using paired-end RAD data, so as to have sufficient flanking region around the SNPs to conduct marker design. To do this, we used a simple pipeline which catalogued the Read 1 data in stacks and then used the assembler cortex_var to conduct de novo assembly and genotyping of multiple samples using the Read 2 data. The analysis of around 1.1 billion short reads of sequence data was reduced to a set of 2579 high-quality candidate SNP markers that were polymorphic in Norwegian and Bavarian beaver. Both laboratory validation of a subset of eight of the SNPs (1.3% error) and internal validation by confirming patterns of Mendelian inheritance in a family group (0.9% error) confirmed the success of this approach