FLASH (Fast Length Adjustment of SHort reads) is an accurate and fast tool
to merge paired-end reads that were generated from DNA fragments whose
lengths are shorter than twice the length of reads. Merged read pairs result
in unpaired longer reads, which are generally more desired in genome
assembly and genome analysis processes.
Briefly, the FLASH algorithm considers all possible overlaps at or above a
minimum length between the reads in a pair and chooses the overlap that
results in the lowest mismatch density (proportion of mismatched bases in
the overlapped region). Ties between multiple overlaps are broken by
considering quality scores at mismatch sites. When building the merged
sequence, FLASH computes a consensus sequence in the overlapped region.
More details can be found in the original publication
Limitations of FLASH include:
- FLASH cannot merge paired-end reads that do not overlap.
- FLASH is not designed for data that has a significant amount of indel
errors (such as Sanger sequencing data). It is best suited for Illumina
On UNIX-compatible systems, including GNU/Linux and Mac OS X, you must compile
FLASH from source. The only dependency, other than functions that are expected
to be available in the C library, is the zlib data compression library. To
install FLASH, download the tarball, untar it, and compile the code using the
$ tar xzf FLASH-1.2.11.tar.gz
$ cd FLASH-1.2.11
The executable file that is produced is named 'flash'. To run it from the
command line you must copy it to a location on your $PATH variable, or else run
it with a path including a directory, such as "./flash".
FLASH also runs on Windows, and you can compile it on Windows using MinGW.
However, for convenience you may instead download a standalone Windows binary
from the SourceForge page (https://sourceforge.net/projects/flashpage/).
Please compile FLASH and run `flash --help' to see command-line usage
information and information about input/output files.
By default, FLASH uses multiple threads. There are "combiner" threads that do
the actual read combining, as well as up to 5 threads that are used for I/O (up
to 2 readers, up to 3 writers). The default number of combiner threads is the
number of processors; however, it can be adjusted with the -t option (long
When multiple combiner threads are used, the order of the combined and
uncombined reads in the output files will be nondeterministic. If you need to
enforce that the output reads appear in the same order as the input, you must
Since the FLASH algorithm considers each read pair independently, FLASH will, by
default, process read pairs in parallel. FLASH v1.2.9 and later also make use
of vector instructions available on modern x86 CPUs. Consequently, FLASH works
quite fast, even with low-cost computing resources. As an example, we ran FLASH
v1.2.9 on a laptop with a dual-core 2.3 GHz AMD x86_64 processor and it
processed one million 101-bp read pairs in 11.6 seconds with the default
parameters. Less than 2 MB of memory was used. Actual timing results will
vary, but they will depend primarily on the number of CPUs available, the speed
of each CPU, and on the I/O speed of reading the input files and writing the
output files. FLASH is designed to be scalable to dozens of processors,
although its speed may be limited by I/O in such cases.
With reads' error rate of 1% or less, FLASH processes over 99% of read pairs
correctly. With error rate of 2%, FLASH processes over 98% of read pairs
correctly when default parameters are used. With more aggressive parameters
(i.e., -x 0.35), FLASH processes over 90% of read pairs correctly even when the
error rate is 5%.
Title: FLASH: fast length adjustment of short reads to improve genome assemblies
Authors: Tanja Magoč and Steven L. Salzberg
FLASH is released under the GNU General Public License Version 3 or later (see
Send an e-mail to email@example.com
Other versions are available from the SourceForge page: