View Single Post
Old 08-21-2010, 03:40 PM   #27
robs
Senior Member
 
Location: San Diego, CA

Join Date: May 2010
Posts: 116
Default

Quote:
Originally Posted by aleferna View Post
The first time I ran BWA with the long aligner I didn't realize that there was a short/long option and since I have both in my library I was very disappointed of BWA. I started testing algorithm after algorithm and finally reviewed BWA again. This time I made a small script that will just join 2 sam files, one for the small aligner and one from the long aligner. It will choose the alignment from the short aligner if it cannot find it in the long aligner, this was the winning combination.

I've mentioned this chart in another thread, but here you can see that BWA is the only one that can cover the full range of read sizes in 454 datasets (or in 100bp solexa data after you remove the pair end adapters!)

http://www.nada.kth.se/~afer/benchmark.jpeg

Moreover, I know using the Z=100 seems a bit of an overkill but with 454 data and a decent computer BWA will take just a few minutes and I did measure Z=1,10,25,50,100,250 and even 500. Z = 100 seems to be the peak, after this I cannot squeeze any specificity out of the algorithm, but you do see a change from Z=10 to Z=100.
Looking at your chart, you actually get better sensitivity for longer reads with low error rates using the default settings instead of using Z=100. Any idea what causes a higher Z-best value to result in lower sensitivity?
robs is offline   Reply With Quote