Comparing bwa with shrimp

[jmj1091]

Hi,

I am hoping to do a direct comparison of the results returned by each of bwa and shrimp. I would like to tune the parameters of bwa's aln algorithm to match the default parameters of shrimp, but I am having trouble doing so. In particular, I do not see an option for controlling the match score, and I am also not sure what kinds of limits (if any) shrimp puts on the number of gap opens/extensions. I do see that bwa's dbwtsw algorithm has a parameter for setting the match score, but the reads are 35bp each, so I am not sure if that function is the way to go. Is it possible to do what I am proposing?

Thanks!

[nilshomer]

Hi,

I am hoping to do a direct comparison of the results returned by each of bwa and shrimp. I would like to tune the parameters of bwa's aln algorithm to match the default parameters of shrimp, but I am having trouble doing so. In particular, I do not see an option for controlling the match score, and I am also not sure what kinds of limits (if any) shrimp puts on the number of gap opens/extensions. I do see that bwa's dbwtsw algorithm has a parameter for setting the match score, but the reads are 35bp each, so I am not sure if that function is the way to go. Is it possible to do what I am proposing?

Thanks!

SHRiMP performs fully gapped local alignment (Smith Waterman), and BWA does not. BWA was designed for fast alignment of high-quality short reads (<2% error rates) and not necessarily for flexibility (ability to align under any scheme). This is true of most aligners. BWA's dbwtsw algorithm is mean for longer reads (hundreds of base pairs). You'll find that on high-quality data, BWA and SHRiMP have similar power, with SHRiMP being able to find a larger fraction of longer indels (>5bp), even with using log gaps and the -e parameter in BWA. The biggest difference will be the timing (the SHRiMP authors are working on this).

Have you considered other alignment methods? For example, Bowtie is also extremely fast and good with high-quality data (although no indels). BFAST is good with SOLiD data as well as finding longer indels. MAQ is the typical benchmark that most aligners use. The list goes on...

Nils

[jmj1091]

Thanks for your response, Nils. Yes, I have considered other alignment methods, including bowtie. My goal is to find another alignment tool (in addition to shrimp) that allows for indels, and to compare it with shrimp in terms of mapping quality and speed. I have not yet used BFAST, but I am in the process of creating an index so I can test it.

Do you know which other alignment tools might be suitable to use? I've heard that novoalign supports indels, but I haven't used it yet. I'm primarily looking for something that is inherently faster than shrimp, and/or has built-in support for multi-threading/parallelizing, while still having functionality for dealing with indels.

[nilshomer]

Thanks for your response, Nils. Yes, I have considered other alignment methods, including bowtie. My goal is to find another alignment tool (in addition to shrimp) that allows for indels, and to compare it with shrimp in terms of mapping quality and speed. I have not yet used BFAST, but I am in the process of creating an index so I can test it.

Do you know which other alignment tools might be suitable to use? I've heard that novoalign supports indels, but I haven't used it yet. I'm primarily looking for something that is inherently faster than shrimp, and/or has built-in support for multi-threading/parallelizing, while still having functionality for dealing with indels.

Being the author of BFAST, and having done many, many simulations myself, my advice is to use BFAST (which is faster and more sensitive than SHRiMP). In a week or two I am going to release a version that will work with minimal RAM requirements (<4GB). So many algorithms do not use Smith Waterman (gapped local alignment) as their finishing step, and as such should be avoided. To my knowledge, only BFAST and SHRiMP do this finishing step. All other methods (heuristics) to find indels (typically involving re-alignment) may cause weird artifacts as well as decreased power as to improve speed. Feel free to PM me and we can continue this discussion offline.

Nils