Has anyone satisfactory first-hand experience used Sniper for SNP calling? It claims to "fully account for sequencing error, template bias, and multi-locus SNP combinations, maintaining high sensitivity and specificity under a broad range of conditions."
There are indeed some interesting bits in the paper. Nonetheless, it seems that the authors did not control the SNPs around indels, which I believe is a major source of errors in SNP calling and has been known for a couple of years at least. Using bowtie makes this worse. They showed that Sniper performed well on simulated data, but I guess they were not simulating indels. I have not read careful enough to know how they evaluated false positives using the 1000g data, but I am surprised that they did not mention ts/tv, the most frequently used measurement. The authors did evaluate the accuracy on Harsimendy et al. data set, but it is target sequencing and quite old. Also, people who are familiar the 1000g SNP calling process would know that 1000g calls are very conservative for a high specificity. I would not be surprised if one could find 20% more SNPs.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6