I hate usearch for being so good



Written by: Richard Smith Last updated: 2013-09-03 18:15:00 +0000

Aligning biological sequences to other biological sequences is the bread and butter of bioinformatics.

By far the most popular software for this purpose is BLAST, the Basic Local Alignment Search Tool. BLAST was published in 1990, and at the time its heuristic algorithm was a huge advance (in computational speed) over the Smith-Waterman algorithm. People have since used it for all manner of sequence alignment tasks.

More recently, alternative aligners have risen to prominence in some specific alignment cases. For example, for aligning short next-generation sequencing reads to a set of longer sequences, bowtie, bwa, SOAP, and friends are all orders of magnitude faster than BLAST, and this is recognised by their usage.

But in the case of plain old alignment of one or more nucleotide or amino acid sequences against a database of one or more nucleotide or amino acid sequences, BLAST still dominates. The problem is, BLAST is SLOW. Really, really slow. And the rate at which we generate sequence data is now really, really fast. BLAST can't keep up.

In fact, if it wasn't for USEARCH, a new aligner that outpaces BLAST by several orders of magnitude on the mundane alignment tasks it's so famous for, most of the work I've done in the first year of my PhD would not have been possible. Some days I'm aligning several thousand assembled contigs to a reference genome, and repeating the process several thousand times over (for example assemblotron might align millions of fairly long sequences into a reference database into a single run). With USEARCH that takes hours. With BLAST it takes weeks or months.

Alt text

Benchmarks show that for nucleotide alignments, USEARCH processes ~300 alignments for every one that BLAST can manage, with no loss in sensitivity. This matches my experience, and in some cases I would say I've had significantly greater speed improvements than this. I'll record some benchmarks of my own next time I can bring myself to fire up BLAST again.

The problem with USEARCH, though, is its license. Only the 32bit version is free, meaning a hard 4GB memory limit. But even worse than that, it's closed source and the method is essentially unpublished (there's a paper, but the details are scant - certainly missing information that would be required to recreate the software from scratch). I can run my alignments with USEARCH until the cows come home, and I can verify the results by comparison with BLAST results. But for every result I don't verify, I can never be sure it's not done something unexpected. I can't check the source code or the mathematical working of the algorithm, so ultimately I don't trust it.

It's worth mentioning that the author of USEARCH, Robert Edgar, is clearly a skilled computational biologist. He also wrote MUSCLE, the multiple sequence aligner that made CLUSTAL look silly when it arrived. And he's not a publically funded researcher - he's an independent researcher and needs to make a living from selling his software - so I don't think he's has any ethical imperative to disclose the inner workings of the product. But that doesn't make the situation any better for users.

To my knowledge there is nothing that comes close to the speed of USEARCH. The world needs an open source aligner with USEARCH's speed and sensitivity.

For now, I'm using USEARCH, but I feel like I'm violating myself by doing so. I hate you USEARCH, for being so damn good and not telling us how.


← Previous Archive Next →

blog comments powered by Disqus