This is an old revision of the document!

# NGS Read Mapping Software on Mogon

As a first introduction into NGS alignment software tools we recommend reading this short blog post. Or in other words: It might be, that the list of supported tools grows and grows, due to your requests, but will never really cover everybody's favorite tool.

Notwithstanding, own benchmarks a first impression can be found in the same blog.

BWA is one mapping tool, particularly to map “low-divergent sequences against a large reference genome”. Modules on Mogon can be found as1):

bio/BWA/<version>

#### The Wrapper Script

To leverage the task from 1 (or a few) samples to be mapped to several in parallel, we provide a wrapper script, which is available as a module:

bio/parallel_BWA

The code is under version management and hosted internally, here.

The wrapper script will submit a job, it is not intended to be just within a SLURM environment, but rather creates one.

Calling parallel_BWA -h will display a help message with all the options, the script provides. Likewise, the call parallel_BWA –credits will display credits and a version history.

parallel_BWA [options] <referencedir> <inputdir> Limitations: • The wrapper recognizes FASTQ files with suffixes “*.gz”, “*.fastq” or “*.fq” and will allways assume FASTQ files (compressed or uncompressed). • The number of processes (and therefore nodes) is limited to the number of samples. • The wrapper only works for paired end sequencing data, where the file tuples are designated with the following strings “_1” and “_2” or “_R1” and “_R2”, respectively. • BWA does not scale well to big data. It is better to split input to chuncks of ~1GB (take this with a grain of salt: there are not scaling tests, yet) • BWA does not scale well beyond a NUMA block (8 threads on Mogon I) • There are only a few options, as internally the wrapper calls bwa mem (or bwa aln in the single end case) and only sets up a few things to yield performance. About Arguments: • referencedir needs to be the (relative) path to a directory containing an indexed BWA reference • inputdir needs to be a (relative) path to a directory containing all inputs. Subdirectories and files containing the string unpaired are ignored; this is to support preprocessing with the trimmomatic module. The options: • parallel_BWA attempts to deduce your SLURM account. This may fail, in which case -A, –account needs to be supplied. • -N,–nodes allows to reserve more than 1 node (the default). This may speed up the screening; see the limitations above. • -d,–dependency, list of comma separated jobids, the job will wait for to finish • -l,–runlimit, this defaults to 300 minutes. • -p,–partition, the default is nodeshort or parallel on Mogon2, no smp-partition should be choosen. • -t,–threads, BWA can work in parallel. Please consult the manual. The default is 8. • -o,–outdir output directory path (default is the current working directory) • –single (no arguments) to evaluate single end data • –args to supply additional flags, e. g. –args=“-l 1024 -n 0.02” for BWA - note the quotation marks, they are necessary. Output: • Per input tuple (paired sequencing data, only) a BAM file with the prefix of the input will be written. In the case of single end data, there will be one output per input, only. Barracuda is a GPU-accelerated implementation of BWA and can be found on Mogon as the module bio/barracuda It does not support bwa mem … but rather leverages bwa aln … to GPUs. See below for a wrapper script to ease your workflow. RazerS 3 as yara is part of the seqan modules: bio/SeqAn/<version> You can find a wrapper to ease your workflow, below, eventually 2). Bowtie2 is a well known read aligner with a focus on gapped alignments. #### The Wrapper Script segemehl seems to be a pretty good alignment tool, mentioned here, due to the blog which is cited below. There will be no wrapper script for segemehl: If this comparison bears any truth, the software might be really good. But also pretty memory hungry. And several tens GB / core is just too much. If you want to try segemehl, be sure to write your own wrapper script (perhaps stage-in the reference to a local scratch, not the ramdisk) and reserve sufficient memory. Be aware that you will be accounted for the prolonged run time and memory. yara is a mapping tool with “with approximate seeds and multiple backtracking”. It is available within the modules bio/SeqAn/<version> You can find a wrapper to ease your workflow, below. Most mapping tools adhere to this paradigm: They work on a reference (directory). They are, therefore, easily wrapped, such that the reference can be staged-in to a node-local directory (e.g. a ramdisk) in order to avoid random I/O (and consequently prolonged run times) on the parallel file system. Whilst adhering to the same paradigm, mentioned above, barracuda is the only read-mapping software supported, which works on GPUs3). This is different and peculiar in its setup and merits a separate module: To leverage the task from 1 (or a few) samples to be mapped to several in parallel, we provide a wrapper script, which is available as a module: bio/parallel_Barracuda Calling parallel_Barracuda -h will display a help message with all the options, the script provides. Likewise, the call parallel_Barracuda –credits will display credits and a version history. The script, after loading the module, can then be run like:  parallel_Barracuda [options] <referencedir> <inputdir>

Limitations:

• See the parallel_BWA wrapper
• Also: The script will only use the m2_gpu partition and therefore needs an account with the m2_ prefix.

• referencedir needs to be the (relative) path to a directory containing an indexed BWA reference. No symbolic links are allowed.
• inputdir needs to be a (relative) path to a directory containing all inputs. Subdirectories and files containing the string unpaired are ignored; this is to support preprocessing with the trimmomatic module.

The options:

• parallel_BWA attempts to deduce your SLURM account. This may fail, in which case -A, –account needs to be supplied.
• -d,–dependency, list of comma separated jobids, the job will wait for to finish
• -l,–runlimit, this defaults to 300 minutes.
• -o,–outdir output directory path (default is the current working directory)

Output:

• Per input tuple (paired sequencing data, only) a BAM file with the prefix of the input will be written. In the case of single end data, there will be one output per input, only.

This part needs some more time to be finished ….

1)