>One thing I really enjoy about this blog is that it seems to encourage communication between people in the 2nd Generation Genomics field. In particular, I get to hear from a lot of people that I otherwise would never have the privilege of meeting. After my posting on the QuEST software for ChIP-Seq the other day, I got a reply from Anton Valouev, the first author on the QuEST paper. (Who’s name I accidentally misspelled in the original post. Sorry – it’s now been fixed!)
He was kind enough to point out some of my mistakes, and to point me to the wikipedia entry on Kernel Density Estimation (KDE), which I had misunderstood as a windowing function. Anyhow, not only was he willing to teach me some science, he was also kind enough to let me print his reply here, which rebuts several of my points very well, and clearly points out the significance of the paper, which I had understated.
Here’s Anton Valouev’s reply:
my name is Anton Valouev and I really enjoyed your blog post featuring QuEST. I wanted to point out a couple of things about your blog post that I wanted to clarify.
Although kernel density estimation uses bandwidth, it is not strictly a windowed approach since the points beyond the bandwidth window contribute to the estimation (see the formula in the methods section of our paper). KDE is an extremely powerful approach and surpasses windowed counts in every single respect except perhaps in simplicity of calculations. Wikipedia provides good reading on KDE approaches.
The novelty of the paper is in the software itself which appears to outperform the Wold lab peak caller (featured in Joh[n]son et al, although we do not demonstrate this in the paper as you have noted).
For these reasons, QuEST is currently adopted as a default peak caller by the ENCODE consortium. Yes, the analysis of the data is somewhat similar to Johnson et al Science paper, but the software is completely new which is the main point of the paper. Another new result is in direct detection of the ELK4 motif in the SRF data which hasn’t been shown before. Obviously motif finding is not new, but this particular result is.
The code dealing with controls is in the peak_caller.cpp lines 248-256, so this is not an unreleased feature. Also, I’d really appreciate if you can change my last name to correct spelling.
Good luck with your FindPeaks software, ChIP-Seq is seeing its early days, so much work is still to be done in this area.
Very best, Anton.