Burrows-Wheeler Aligner (BWA) is an efficient program that aligns relatively short nucleotide sequences against a long reference sequence such as the human genome. This app runs BWA to map letterspace reads to a reference genome and produce mappings.

This app accepts one or more gtable object of type Reads. The reads in all of these tables will be mapped to the genome and output in a single gtable of type Mappings. (See http://wiki.dnanexus.com/Types/Mappings for more information on the Mappings type.) Mappings objects can be then used as inputs to certain variation calling apps, mappings QC apps, etc.; they can also be visualized in the genome browser.

BWA requires a special processing on the genome, called indexing; this processing can take several hours for long genomes. If the genome given in the input is not indexed for BWA, the app will automatically index it and include an indexed version in the output, for future use. When possible, please run this app with an indexed genome to avoid re-indexing. DNAnexus provides several pre-indexed genomes in the 'Reference Genomes' public project.

BWA implements two different algorithms, both based on the Burrows-Wheeler Transform (BWT). The first algorithm, called 'aln', is designed for short query sequences up to ~200bp with low error rate (<3%). It does gapped global alignment with respect to the queries, supports paired-end reads, and is one of the fastest short read alignment algorithms to date while also visiting suboptimal hits. The second algorithm, called 'bwasw', is designed for long reads with more errors. It performs heuristic Smith-Waterman-like alignment to find high-scoring local hits (and thus chimera). On low-error short queries, 'bwasw' is slower and less accurate than the first algorithm, but on long queries, it is better. Using a value of 'auto' will automatically choose a suitable algorithm based on the length of the reads in the input (however, this is currently not implemented, and using 'auto' will lead to the 'aln' algorithm always being chosen).

This app parallelizes itself by dividing the input into chunks of a certain size (a certain number of reads), and mapping each chunk individually. Lower chunk sizes lead to higher levels of parallelization, reducing the wall-clock time that one has to wait for the app to finish. However, lower chunks sizes may also increase the cost of running the app in the cloud, as they lead to a higher number of chunks, each of which adds a constant processing overhead. The default value is 25 million reads per chunk; DNAnexus suggests caution when experimenting with this parameter.

Users familiar with the BWA executable can directly manipulate the parameters that are used for the bwa aln, bwa samse, bwa sampe and bwa bwasw calls. These parameters are: aln_n, aln_o, aln_e, aln_i, aln_d, aln_l, aln_k, aln_m, aln_M, aln_O, aln_E, aln_R, aln_q, aln_N, sampe_a, sampe_o, sampe_n, sampe_N, sampe_c, sampe_s, samse_n, sw_a, sw_b, sw_q, sw_r, sw_w, sw_m, sw_T, sw_c, sw_z, sw_s, sw_N. Each one of this parameters directly correspond to the respective command-line argument, e.g. aln_o corresponds to the -o option of bwa aln (maximum number of gap opens). Certain options, such as the -t option of bwa aln, are not exposed to users because they are set by the app, based on the kind of cloud environment that the app runs on.