nimbliner - nimble aligner

Fast and lightweight read aligner (experimental)

Nimbliner uses Bloom filters instead of suffix arrays as reference which incurs the cost close to n in the size of reference sequence (instead of 2-4n for suffix arrays or BWT). It also does not need to perform a full alignment shaving off a lot of the computational cost. Nimbliner does not yet produce cigar strings, but there is no reason why it would not be able to.

Installation

You can build a docker image and then run nimbliner within the container:

<clone the repo>
docker build -t nimbliner-dev:0.1 -f docker/Dockerfile docker/
# this will produce all_kmers.txt and anchors.txt in the current directory
docker run -v `pwd`:/nimbliner nimbliner-dev:0.1 indexer 20 <path to your reference, single fasta file>
docker run -v `pwd`:/nimbliner nimbliner-dev:0.1 mapper 20 <path to your reads, single fasta file> all_kmers.txt anchors.txt

Running benchmarks

Benchmarking is set up with [Snakemake](). To run the benchmarks, you can do:

cd benchmarking
snakemake --configfile experiments.json run_pipeline

Other ways

You can compile from source. The dependencies are liffb and TCLAP. You may need to set LD_LIBRARY_PATH (or DYLD_LIBRARY_PATH for MacOS) to /usr/local/lib since libbf installs there by default.

Niceties

You can generate synthetic reads w/ mismatches and indels. For example, to sample a million reads from the chromosome w/ 1.5% error rate, do:

docker run -v `pwd`:/nimbliner nimbliner-dev:0.1 sample 20 1000000 chromo.fa 1.5 > sampled_reads.fa

TODO

• [ ] provide benchmark data
• [ ] prepare indices for the whole human genome
• [ ] integrate with TravisCI

Comparisons:

• DALIGNER
• STAR
• BWA
• RapMap (speed-only, RapMap does not generate full alignments)