cutadapt removes adapter sequences from DNA high-throughput sequencing data. This is usually necessary when the read length of the machine is longer than the molecule that is sequenced, such as in microRNA data.
cutadapt is implemented in Python. It comes with an extension module written in Cython that implements the alignment algorithm.
See http://code.google.com/p/cutadapt/ . Please use the Google code issue tracker for bug reports and feature requests.
(This is the MIT license.)
Copyright (c) 2010-2013 Marcel Martin marcel.martin@tu-dortmund.de
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
cutadapt needs Python 2.6 or later, including Python 3 (tested with Python 3.1.2 and 3.2.2). Using Python 2.7 is recommended since it works fastest. For installation from sources, a C compiler needs to be installed. The program has been developed and tested on Ubuntu and OpenSuSE.
Replace "python" with "python3" in the following lines to install the Python 3 version.
python setup.py build
python setup.py install
If you get an error about a missing "Python.h" file, then make sure that the python-dev package is installed (or python3-dev for Python 3).
Build the C extension module (you can try to skip this step -- a compiled version of the module for Linux x86 is already included):
python setup.py build_ext -i
Then simply run the script from where it is, similar to this:
bin/cutadapt --help
If you get any errors, first try to explicitly request a specific Python version by running cutadapt like this:
python2.7 bin/cutadapt --help
If you want to use cutadapt within the web-based Galaxy platform (http://galaxy.psu.edu/), please see the README file in the galaxy/ subfolder. Galaxy support was contributed by Lance Parsons.
Please also see the command-line help: cutadapt --help
The basic command-line for cutadapt looks like this:
cutadapt -a AACCGGTT input.fastq > output.fastq
The adapter sequence is given with the -a option. Replace
AACCGGTT with your actual adapter sequence.
input.fastq is a file with reads. The result will be written
to standard output. Use redirection with > (or the -o option) to
write the output to a file.
cutadapt also writes a report after it has finished processing the
reads. If you use -o, the report is sent to standard output and to
stderr otherwise.
By default, the output file contains all reads, even those
that did not contain an adapter. (See also the --discard option.)
The following examples refer to basespace reads. See the "Colorspace" section on how to use cutadapt with SOLiD reads.
Assuming your sequencing data is available as a FASTQ file, use this command line:
cutadapt -a ADAPTER-SEQUENCE input.fastq > output.fastq
gz-compressed input is supported:
cutadapt -a ADAPTER-SEQUENCE input.fastq.gz > output.fastq
gz-compressed output is also supported, but the -o parameter (output file) needs to be used (gzip compression is auto-detected by looking at the file name):
cutadapt -a ADAPTER-SEQUENCE -o output.fastq.gz input.fastq.gz
If your Python installation includes support for bzip2 compression, then
bzip2-compressed files are also supported and recognized by their extension
.bz2.
You can give names to the adapters. The names are shown in addition to the sequences themselves in the statistics overview when the program has finished trimming the reads. You can use it like this:
cutadapt -a My_adapter=ACGTAA input.fastq > output.fastq
Here, the actual adapter sequence is ACGTAA and the name assigned
to it is My_adapter.
Cut an adapter from reads given in a FASTA file. Try to remove an adapter three times (this is usually not needed), use the default error rate of 10%, write result to output.fa:
cutadapt -n 3 -a TGAGACACGCAACAGGGGAAAGGCAAGGCACACAGGGGATAGG input.fa > output.fa
As many adapters as desired can be given to the program by using the -a, -b or -g
in any combination, for example, five -a adapters and two -g adapters. All
adapters will be searched for, but only the best matching one will be trimmed
from each read (but see the --times option).
cutadapt -b TGAGACACGCA -g AGGCACACAGGG input.fastq > output.fastq
The -q (or --trim-qualities) parameter can be used to trim low-quality ends
from reads before adapter removal. For this to work correctly, the quality
values must be encoded as ascii(phred quality + 33). If they are encoded as
ascii(phred quality + 64), you need to add --quality-base=64 to the command line.
The trimming algorithm is the same as the one used by BWA. That is: Subtract the given cutoff from all qualities; compute partial sums from all indices to the end of the sequence; cut sequence at the index at which the sum is minimal.
Cutadapt supports paired-end trimming, but currently two passes over the data are required.
Assume the input is in reads.1.fastq and reads.2.fastq and that
ADAPTER_FWD should be trimmed from the forward reads (first file) and
ADAPTER_REV from the second reverse reads (second file). There are
two cases.
If you do not use any of the options that discard reads,
such as --discard, --minimum-length or --maximum-length, then
run cutadapt on each file separately:
cutadapt -a ADAPTER_FWD -o trimmed.1.fastq reads1.fastq
cutadapt -a ADAPTER_REV -o trimmed.2.fastq reads2.fastq
If you use one of the read-discarding options, then the --paired-output
option is needed to keep the two files synchronized. First trim the forward read,
writing output to temporary files:
cutadapt -a ADAPTER_FWD --minimum-length 20 --paired-output tmp.2.fastq -o tmp.1.fastq reads.1.fastq reads.2.fastq
Then trim the reverse read, using the temporary files as input:
cutadapt -a ADAPTER_REV --minimum-length 20 --paired-output trimmed.1.fastq -o trimmed.2.fastq tmp.1.fastq tmp.2.fastq
Finally, remove the temporary files:
rm tmp.1.fastq tmp.2.fastq
The adapters that should be trimmed from Illumina TruSeq are the following.
Trim read 1 with A + the “TruSeq Indexed Adapter”:
cutadapt -a AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC reads.1.fastq > trimmed.1.fastq
Trim read 2 with the reverse complement of the ”TruSeq Universal Adapter”:
cutadapt -a AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGTAGATCTCGGTGGTCGCCGTATCATT reads.2.fastq > trimmed.2.fastq
See also the section about paired-end adapter trimming above.
The adapter sequences can be found in the document Illumina TruSeq Adapters De-Mystified.
These are some 454 adapters: A1: 5'- TCCATCTCATCCCTGCGTGTCCCATCTGTTCCCTCCCTGTCTCA A2: 5'- TGAGACAGGGAGGGAACAGATGGGACACGCAGGGATGAGATGGA B1: 5'- CCTATCCCCTGTGTGCCTTGCCTATCCCCTGTTGCGTGTCTCA B2: 5'- TGAGACACGCAACAGGGGAAAGGCAAGGCACACAGGGGATAGG
This is an AB SOLiD adapter (in color space) used in the SREK protocol: 330201030313112312
cutadapt uses a modified semi-global alignment algorithm. For speed, the
algorithm is implemented as a Python extension module in calignmodule.c.
Cutadapt’s processing speed is currently not dominated by the alignment algorithm, but by parsing the input and writing the output.
Cutadapt correctly deals with partial adapter matches, and also with any trailing
sequences after the adapter. As an example, suppose your adapter sequence is
"ADAPTER" (specified via the -a or --adapter command-line parameter).
If you have these input sequences:
MYSEQUENCEADAPTER
MYSEQUENCEADAP
MYSEQUENCEADAPTERSOMETHINGELSE
All of them will be trimmed to "MYSEQUENCE". If the sequence starts with an adapter, like this:
ADAPTERSOMETHING
It will be empty after trimming.
When the allowed error rate is sufficiently high (set with parameter -e), errors in
the adapter sequence are allowed. For example, ADABTER (1 mismatch), ADAPTR (1 deletion),
and ADAPPTER (1 insertion) will all be recognized if the error rate is set to 0.15.
If you specify an adapter with the -g (--front) parameter, the adapter may
overlap the beginning of the read or occur anywhere within it. If it appears
within the read, the sequence that precedes it will also be trimmed in addition
to the adapter. For example, with -g ADAPTER, these sequences:
HELLOADAPTERTHERE
APTERTHERE
will both be trimmed to THERE. To avoid this, you can prefix the adapter with the
character ^. This will restrict the search, forcing the adapter to be a prefix
of the read. With -g ^ADAPTER, only reads like this will be trimmed:
ADAPTERHELLO
Cutadapt assumes that any adapter specified via the -a (or --adapter) parameter
was ligated to the 3' end of the sequence. This is the correct assumption for
at least the SOLiD and Illumina small RNA protocols and probably others.
The assumption is enforced by the alignment algorithm, which only finds the adapter
when its starting position is within the read. In other words, the 5' base of
the adapter must appear within the read. The adapter and all bases following
it are remved.
If, on the other hand, your adapter can also be ligated to the 5' end (on
purpose or by accident), you should tell cutadapt so by using the -b (or
--anywhere) parameter. It will then use a slightly different alignment algorithm
(so-called semiglobal alignment), which allows any type of overlap between the
adapter and the sequence. In particular, the adapter may appear only partially
in the beginning of the read, like this:
PTERMYSEQUENCE
The decision which part of the read to remove is made as follows: If there is at least one base before the found adapter, then the adapter is considered to be a 3' adapter and the adapter itself and everything following it is removed. Otherwise, the adapter is considered to be a 5' adapter and it is removed from the read.
Here are some examples, which may make this clearer (left: read, right: trimmed read):
MYSEQUENCEADAPTER -> MYSEQUENCE (3' adapter)
MADAPTER -> M (3' adapter)
ADAPTERMYSEQUENCE -> MYSEQUENCE (5' adapter)
PTERMYSEQUENCE -> MYSEQUENCE (5' adapter)
The regular algorithm (-a) would trim the first two examples in the same way,
but trim the third to an empty sequence and trim the fourth not at all.
The -b parameter currently does not work with color space data.
After every run, cutadapt prints out per-adapter statistics. The output starts with something like this:
Adapter 'ACGTACGTACGTTAGCTAGC', length 20, was trimmed 2402 times.
The meaning of this should be obvious.
The next piece of information is this:
No. of allowed errors:
0-9 bp: 0; 10-19 bp: 1; 20 bp: 2
The adapter has, as was conveniently shown above, a length of 20 characters. We are using the default error rate of 0.1. What this implies is shown above: Matches up to a length of 9 bp are allowed to have no errors. Matches of lengths 10-19 bp are allowd to have 1 error and matches of length 20 can have 2 errors.
Finally, a table is output that gives more detailed information about the lengths of the removed sequences. The following is only an excerpt; some rows are left out:
Overview of removed sequences
length count expect max.err error counts
3 140 156.2 0 140
4 57 39.1 0 57
5 50 9.8 0 50
6 35 2.4 0 35
...
100 397 0.0 3 358 36 3
The first row tells us the following: Three bases were removed in 140 reads; randomly, one would expect this to occur 156.2 times; the maximum number of errors at that match length is 0 (this is actually redundant since we know already that no errors are allowed at lengths 0-9bp).
The last column shows the number of reads that had 0, 1, 2 ... errors. In the last row, for example, 358 reads matched the adapter with zero errors, 36 with 1 error, and 3 matched with 2 errors.
The "expect" column gives only a rough estimate of the number of sequences that is expected to match randomly (it assumes a GC content of 50%, for example), but it can help to estimate whether the matches that were found are true adapter matches or if they are due to chance. At lengths 6, for example, only 2.4 reads are expected, but 35 do match, which hints that most of these matches are due to actual adapters.
Note that the "length" column refers to the length of the removed sequence. That is, the actual length of the match in the above row at length 100 is 20 since that is the adapter length. Assuming the read length is 100, the adapter was found in the beginning of 397 reads and therefore those reads were trimmed to a length of zero.
The table may also be useful in case the given adapter sequence contains an error. In that case, it may look like this:
...
length count expect max.err error counts
10 53 0.0 1 51 2
11 45 0.0 1 42 3
12 51 0.0 1 48 3
13 39 0.0 1 0 39
14 40 0.0 1 0 40
15 36 0.0 1 0 36
...
We can see that no matches longer than 12 have zero errors. In this case, it indicates that the 13th base of the given adapter sequence is incorrect.
When the --info-file command-line parameter is given, detailed information
about the found adapters is written to the given file. The output is a
tab-separated text file. Each line corresponds to one read of the input file.
The columns are:
- Read name
- Number of errors
- 0-based start coordinate of the adapter match
- 0-based end coordinate of the adapter match
- Sequence of the read before the adapter match
- Sequence of the read that was matched to the adapter
- Sequence of the read after the adapter match
The concatenation of the last three fields yields the full read sequence.
If no adapter was found, the format is as follows:
- Read name
- The value -1
- The read sequence
The option is currently experimental and additional columns may be
added in the future. Also, in the current version, when the --times
option is set to a value other than 1 (the default value), multiple
lines are written to the info file for each read.
Cutadapt currently does not support using a configuration file in which, for example, a list of adapters can be specified. If you have many adapters that you want to seach for and want to avoid typing all of them on the command line, then you can use so-called "command substitution" of your Unix shell. With Bash, this works as follows.
First, create a configuration file cutadapt.conf that contains lines like this:
-a AACCGGTT
-a GTAATAACCGGTT
-e 0.05
The file may contain line breaks (they will be replaced by spaces).
Then run cutadapt like this:
cutadapt $(<cutadapt.conf) input.fastq > output.fastq
The Bash shell will replace the $(<...) with the content of the given file.
Cutadapt was designed to work with colorspace reads from the ABi SOLiD sequencer.
Colorspace trimming is activated by the --colorspace option (or use -c for short).
The input reads can be given either:
- in a FASTA file
- in a FASTQ file
- in a
.csfastaand a.qualfile (this is the native SOLiD format).
In all cases, the colors must be represented by the characters 0, 1, 2, 3.
Example input files are in the cutadapt distribution at tests/data/solid.*.
The .csfasta/.qual file format is automatically assumed if two input
files are given to cutadapt.
In colorspace mode, the adapter sequences given to the -a, -b and -g options
can be given both as colors or as nucleotides. In the latter case, they will
simply be converted automatically. For example, to trim an adapter from
solid.csfasta and solid.qual, use this command-line:
cutadapt -c -a CGCCTTGGCCGTACAGCAG solid.csfasta solid.qual > output.fastq
In case you know the colorspace adapter sequence, you can also write 330201030313112312
instead of CGCCTTGGCCGTACAGCAG and the result is the same.
The read mappers MAQ and BWA (and possibly others) need their colorspace
input reads to be in a so-called "double encoding". This simply means that
they cannot deal with the characters 0, 1, 2, 3 in the reads, but require
that the letters A, C, G, T be used for colors. For example, the colorspace
sequence 0011321 would be AACCTGC in double-encoded form. This is not
the same as conversion to basespace! The read is still in colorspace, only
letters are used instead of digits. If that sounds confusing, that is
because it is.
Note that MAQ is unmaintained and should not be used in new projects.
BWA’s colorspace support was dropped in versions more recent than 0.5.9, but that version works well.
When you want to trim reads that will be mapped with BWA or MAQ, you can
use the --bwa option, which enables colorspace mode (-c), double-encoding
(-d) and primer trimming (-t), all of which are required for BWA, in
addition to some other useful options.
There is also the --maq option, which is simply another name for the
--bwa option.
To cut an adapter from SOLiD data given in solid.csfasta and solid.qual,
to produce MAQ- and BWA-compatible output, allow the default of 10% errors
and write the resulting FASTQ file to output.fastq:
cutadapt --bwa -a CGCCTTGGCCGTACAGCAG solid.csfasta solid.qual > output.fastq
Instead of redirecting standard output with >, the -o option can be used. This
also shows that you can give the adapter in colorspace and how to use a different
error rate:
cutadapt --bwa -e 0.15 -a 330201030313112312 -o output.fastq solid.csfasta solid.qual
This does the same as above, but produces BFAST-compatible output, strips the _F3 suffix from read names and adds the prefix "abc:" to them:
cutadapt -c -e 0.15 -a 330201030313112312 -x abc: --strip-f3 solid.csfasta solid.qual > output.fastq
Quality values of colorspace reads are sometimes negative. Bowtie gets confused and prints this message:
Encountered a space parsing the quality string for read xyz To avoid this problem, use the
--zero-capoption (or the short version-z), which converts negative quality values to zero. Since BWA has a similar problem (it crashes) the option is automatically enabled when--bwais used.
- Add the
--too-long-outputoption. - Add the
--no-trimoption, contributed by Dave Lawrence. - Port handwritten C alignment module to Cython.
- Fix the
--rest-fileoption (issue 56) - Slightly speed up alignment of 5' adapters.
- Support bzip2-compressed files.
- At least 25% faster processing of .csfasta/.qual files due to faster parser.
- Between 10% and 30% faster writing of gzip-compressed output files.
- Support 5' adapters in color space, even when no primer trimming is requested.
- Add the `--info-file' option, which has a line for each found adapter.
- Named adapters are possible. Usage:
-a My_Adapter=ACCGTAassigns the name "My_adapter". - Improve alignment algorithm for better poly-A trimming when there are sequencing errors. Previously, not the longest possible poly-A tail would be trimmed.
- James Casbon contributed the
--discard-untrimmedoption.
- Allow to "anchor" 5' adapters (
-g), forcing them to be a prefix of the read. To use this, add the special character^to the beginning of the adapter sequence. - Add the "-N" option, which allows 'N' characters within adapters to match literally.
- Speedup of approx. 25% when reading from .gz files and using Python 2.7.
- Allow to only trim qualities when no adapter is given on the command-line.
- Add a patch by James Casbon:
- include read names (ids) in rest file
- Use nosetest for testing. To run, install nose and run "nosetests".
- When using cutadapt without installing it, you now need to run
bin/cutadaptdue to a new directory layout. - Allow to give a colorspace adapter in basespace (gets automatically converted).
- Allow to search for 5' adapters (those specified with
-g) in colorspace. - Speed up the alignment by a factor of at least 3 by using Ukkonen's algorithm. The total runtime decreases by about 30% in the tested cases.
- allow to deal with colorspace FASTQ files from the SRA that contain a fake additional quality in the beginning (use --format sra-fastq)
- ASCII-encoded quality values were assumed to be encoded as ascii(quality+33).
With the new parameter
--quality-base, this can be changed to ascii(quality+64), as used in some versions of the Illumina pipeline. (Fixes issue 7.) - Allow to specify that adapters were ligated to the 5' end of reads. This change is based on a patch contributed by James Casbon.
- Due to cutadapt being published in EMBnet.journal, I found it appropriate to call this release version 1.0. Please see http://journal.embnet.org/index.php/embnetjournal/article/view/200 for the article and I would be glad if you cite it.
- Add Galaxy support, contributed by Lance Parsons.
- Patch by James Casbon: Allow N wildcards in read or adapter or both. Wildcard matching of 'N's in the adapter is always done. If 'N's within reads should also match without counting as error, this needs to be explicitly requested via --match-read-wildcards.
-
Fix issue 20: Make the report go to standard output when
-o/--outputis specified. -
Recognize .fq as an extension for FASTQ files
-
many more unit tests
-
The alignment algorithm has changed. It will now find some adapters that previously were missed. Note that this will produce different output than older cutadapt versions!
Before this change, finding an adapter would work as follows:
- Find an alignment between adapter and read -- longer alignments are better.
- If the number of errors in the alignment (divided by length) is above the maximum error rate, report the adapter as not being found. Sometimes, the long alignment that is found had too many errors, but a shorter alignment would not. The adapter was then incorrectly seen as "not found". The new alignment algorithm checks the error rate while aligning and only reports alignments that do not have too many errors.
- now compatible with Python 3
- Add the
--zero-capoption, which changes negative quality values to zero. This is a workaround to avoid segmentation faults in BWA. The option is now enabled by default when--bwa/--maqis used. - Lots of unit tests added. Run them with
cd tests && ./tests.sh. - Fix issue 16:
--discard-trimmeddid not work. - Allow to override auto-detection of input file format with the new
-f/--formatparameter. This mostly fixes issue 12. - Don't break when input file is empty.
- Install a single 'cutadapt' Python package instead of multiple Python modules. This avoids cluttering the global namespace and should lead to less problems with other Python modules. Thanks to Steve Lianoglou for pointing this out to me!
- ignore case (ACGT vs acgt) when comparing the adapter with the read sequence
- .FASTA/.QUAL files (not necessarily color space) can now be read (some 454 software uses this format)
- Move some functions into their own modules
- lots of refactoring: replace the fasta module with a much nicer seqio module.
- allow to input FASTA/FASTQ on standard input (also FASTA/FASTQ is autodetected)
- add
--too-short-outputand--untrimmed-output, based on patch by Paul Ryvkin (thanks!) - add
--maximum-lengthparameter: discard reads longer than a specified length - group options by category in
--helpoutput - add
--length-tagoption. allows to fix read length in FASTA/Q comment lines (e.g.,length=123becomeslength=58after trimming) (requested by Paul Ryvkin) - add
-q/--quality-cutoffoption for trimming low-quality ends (uses the same algorithm as BWA) - some refactoring
- the filename
-is now interpreted as standard in or standard output
- Change default behavior of searching for an adapter: The adapter is now assumed to
be an adapter that has been ligated to the 3' end. This should be the correct behavior
for at least the SOLiD small RNA protocol (SREK) and also for the Illumina protocol.
To get the old behavior, which uses a heuristic to determine whether the adapter was
ligated to the 5' or 3' end and then trimmed the read accordingly, use the new
-b(--anywhere) option. - Clear up how the statistics after processing all reads are printed.
- Fix incorrect statistics. Adapters starting at pos. 0 were correctly trimmed, but not counted.
- Modify scoring scheme: Improves trimming (some reads that should have been trimmed were not). Increases no. of trimmed reads in one of our SOLiD data sets from 36.5 to 37.6%.
- Speed improvements (20% less runtime on my test data set).
- Useful exit codes
- Better error reporting when malformed files are encountered
- Add
--minimum-lengthparameter for discarding reads that are shorter than a specified length after trimming. - Generalize the alignment function a bit. This is preparation for supporting adapters that are specific to either the 5' or 3' end.
- pure Python fallback for alignment function for when the C module cannot be used.
- Support gzipped input and output.
- Print timing information in statistics.
- add --discard option which makes cutadapt discard reads in which an adapter occurs
- (more) correctly deal with multiple adapters: If a long adapter matches with lots of errors, then this could lead to a a shorter adapter matching with few errors getting ignored.
- fix huge memory usage (entire input file was unintentionally read into memory)
- allow FASTQ input
- initial release
- show average error rate
- In color space and probably also for Illumina data, gapped alignment is not necessary
- bzip2 support
- use
str.formatinstead of% - allow to change scores at runtime (using command-line parameters)
- multi-threading
--progress- run pylint, pychecker
- length histogram
- refactor read_sequences (use classes)
- put write_read into a Fast(a|q)Writer class?
- allow .txt input/output
- test on Windows
- check whether input is FASTQ although -f fasta is given
- close on StopIteration
- search for adapters in the order in which they are given on the command line
- more tests for the alignment algorithm