Running LAST in parallel

You can make LAST faster by running it on multiple CPUs / cores. The easiest way is with lastal's -P option:

lastal -P4 my-index queries.fasta > out.maf

This will use 4 parallel threads. If you specify -P0, it will use as many threads as your computer claims it can handle simultaneously.

This works by aligning different query sequences in different threads - so if you only have one query you won't get any parallelization!

Dealing with very long query sequences

lastal aligns one "batch" of queries at a time, so if the batch has only one query you won't get any parallelization. This can be fixed by increasing the batch size, with option -i:

lastal -P4 -i3G my-index queries.fasta > out.maf

This specifies a batch size of 3 gibi-bytes. The downside is that more memory is needed to hold the batch and its alignments.

Dealing with pipelines

If you have a multi-command "pipeline", such as:

lastal -P4 my-index queries.fasta | last-split > out.maf

then the -P option may help, because lastal is often the slowest step, but it would be nice to parallelize the whole thing. Unfortunately, last-split doesn't have a -P option, and even if it did, the pipe between the commands would become a bottleneck.

You can use parallel-fasta and parallel-fastq (which accompany LAST, but require GNU parallel to be installed). These commands read sequence data, split it into blocks (with a whole number of sequences per block), and run the blocks in parallel through any command or pipeline you specify, using all your CPU cores. Here are some examples.

Instead of this:

lastal mydb queries.fa > myalns.maf

try this:

parallel-fasta "lastal mydb" < queries.fa > myalns.maf

Instead of this:

lastal -Q1 -D100 db q.fastq | last-split > out.maf

try this:

parallel-fastq "lastal -Q1 -D100 db | last-split" < q.fastq > out.maf

Instead of this:

zcat queries.fa.gz | lastal mydb > myalns.maf

try this:

zcat queries.fa.gz | parallel-fasta "lastal mydb" > myalns.maf

Notes: