Comparison of Three Methods for SNP Discovery Using Modified ddRADseq Sequencing Data in Allotetraploid Brassica napus

Reduced representation libraries (RRLs) of six rapeseed cultivars were constructed by modified ddRADseq protocol and were sequenced using Illumina technology. In order to compare the efficacy of SNP discovery and the power to distinguish allelic SNPs from homoeologous sequences, two reference-based SNP discovery methods, SOAPsnp and RFAPtools, and one reference-free method, Stacks, were used to mine SNPs among the six cultivars. The Stacks method which based on clustering of sequence tags and a maximum likelihood statistical model discovered a total of 33,316 SNPs, which is the highest. The RFAPtools method used a pseudo-reference sequence assembled from a DH segregating population and detected 28,474 SNPs. The SOAPsnp method used genome sequences of B. rapa and B. oleracea as genome reference sequence and detected 16,286 SNPs, which is the lowest. Of all these SNPs, 11,027 SNPs were called by all three pipelines, and 12,338 SNPs were discovered by two of the three pipelines. SNPs called only by one method usually had lower coverage depth than those called by two or three methods. Compared to RFAPtools, Stacks detected a higher percentage of SNPs distributed to the ends of sequence tags which were prone to be sequencing errors, and it also detected higher percentage of sequence tags with three or four nucleotide variations. And compared to Stacks, RFAPtools could efficiently discover allelic SNPs from the homoeologous sequences, furemore SNPs discovered by RFAPtools were more reliable because base quality was considered.