Trim Galore User Guide V0.4.0
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 12
May 06, 2015
Taking appropriate QC measures for
RRBS-type or other -Seq applications
with Trim Galore!
For all high throughput sequencing applications, we would recommend performing some quality
control on the data, as it can often straight away point you towards the next steps that need to be
taken (e.g. with FastQC). Thorough quality control and taking appropriate steps to remove problems
is vital for the analysis of almost all sequencing applications. This is even more critical for the proper
analysis of RRBS libraries since they are susceptible to a variety of errors or biases that one could
probably get away with in other sequencing applications. In our brief guide to RRBS (RRBS_Guide)
we discuss the following points:
- poor qualities – affect mapping, may lead to incorrect methylation calls and/or mis-mapping
- adapter contamination – may lead to low mapping efficiencies, or, if mapped, may result in
incorrect methylation calls and/or mis-mapping
- positions filled in during end-repair will infer the methylation state of the cytosine used for
the fill-in reaction but not of the true genomic cytosine
- paired-end RRBS libraries (especially with long read length) yield redundant methylation
information if the read pairs overlap
- RRBS libraries with long read lengths suffer more from all of the above due to the short size-
selected fragment size
Poor base call qualities or adapter contamination are however just as relevant for 'normal', i.e. non-
RRBS, libraries.
Adaptive quality and adapter trimming with Trim Galore!
We have tried to implement a method to rid RRBS libraries (or other kinds of sequencing datasets) of
potential problems in one convenient process. For this we have developed a wrapper script
(trim_galore) that makes use of the publically available adapter trimming tool Cutadapt and
FastQC for optional quality control once the trimming process has completed.
Even though Trim Galore! works for any (base space) high throughput dataset (e.g. downloaded
from the SRA) this section describes its use mainly with respect to RRBS libraries.
- In the first step, low-quality base calls are trimmed off from the 3' end of the reads before adapter
removal. This efficiently removes poor quality portions of the reads. Here is an example of a dataset
downloaded from the SRA which was trimmed with a Phred score threshold of 20 (data set
DRR001650_1 from Kobayashi et al., 2012).
before trimming after trimming
In the next step, Cutadapt finds and removes adapter sequences from the 3’ end of reads. If no
sequence was supplied it will attempt to auto-detect the adapter which has been used. For this it
will analyse the first 1 million sequences of the first specified file and attempt to find the first 12 or
13bp of the following standard adapters:
Illumina: AGATCGGAAGAGC
Small RNA: ATGGAATTCTCG
Nextera: CTGTCTCTTATA
If no adapter can be detected within the first 1 million sequences Trim Galore defaults to --illumina.
The auto-detection behaviour can be overruled by specifying an adapter sequence or using --
illumina, --nextera or –small_rna. (Please note the first 13 bp of the standard Illumina paired-end
adapters ('AGATCGGAAGAGC') recognise and removes adapter from most standard libraries,
including the TruSeq and Sanger iTag adapters). To control the stringency of the adapter removal
process one gets to specify the minimum number of required overlap with the adapter sequence;
else it will default to 1. This default setting is extremely stringent, i.e. an overlap with the adapter
sequence of even a single bp is spotted and removed. This may appear unnecessarily harsh;
however, as a reminder adapter contamination may in a bisulfite-Seq setting lead to mis-alignments
and hence incorrect methylation calls, or result in the removal of the sequence as a whole because
of too many mismatches in the alignment process. Tolerating adapter contamination is most likely
detrimental to the results, but we realize that this process may in some cases also remove some
genuine genomic sequence. It is unlikely that the removed bits of sequence would have been
involved in methylation calling anyway (since only the 4th and 5th adapter base would possibly be
involved in methylation calls (for directional libraries that is)), however, it is quite likely that true
adapter contamination – irrespective of its length – would be detrimental for the alignment or
methylation call process, or both.
before trimming after trimming
This example (same dataset as above) shows the dramatic effect of adapter contamination
on the base composition of the analysed library, e.g. the C content rises from ~1% at the
start of reads to around 22% (!) towards the end of reads. Adapter trimming with
Cutadapt gets rid of most signs of adapter contamination efficiently. Note that the sharp
decrease of A at the last position is a result of removing the adapter sequence very
stringently, i.e. even a single trailing A at the end is removed.
- Trim galore! also has an ‘--rrbs’ option for DNA material that was digested with
MspI. In this mode, Trim galore! identifies sequences that were adapter-trimmed and
removes another 2 bp from their 3' end. This is to avoid that the filled-in cytosine position
close to the second MspI site in a sequence is used for methylation calls. Sequences which
were merely trimmed because of poor quality will not be shortened any further.
- Trim Galore! also has a ‘--non_directional’ option, which will screen adapter-
trimmed sequences for the presence of either CAA or CGA at the start of sequences and
clip off the first 2 bases if found. If CAA or CGA are found at the start, no bases will be
trimmed off from the 3’ end even if the sequence had some contaminating adapter
sequence removed (in this case the sequence read likely originated from either the CTOT or
CTOB strand).
- Lastly, since quality and/or adapter trimming may result in very short sequences (sometimes
as short as 0 bp), Trim Galore! can filter trimmed reads based on their sequence length
(default: 20 bp). This is to reduce the size of the output file and to avoid crashes of
alignment programs which require sequences with a certain minimum length.
Note that is not recommended to remove too short sequences if the analysed FastQ file is
one of a pair of paired-end files since this confuses the sequence-by-sequence order of
paired-end reads which is again required by many aligners. For paired-end files, Trim
Galore! has an option ‘--paired’ which runs a paired-end validation on both trimmed
_1 and _2 FastQ files once the trimming has completed. This step removes entire read pairs
if at least one of the two sequences became shorter than a certain threshold. If only one of
the two reads is longer than the set threshold, e.g. when one read has very poor qualities
throughout, this singleton read can be written out to unpaired files (see option
‘--retain_unpaired’) which may be aligned in a single-end manner.
Applying these steps to both self-generated and downloaded data can ensure that you really only
use the high quality portion of the data for alignments and further downstream analyses and
conclusions.
Full list of options for Trim galore!
USAGE:
trim_galore [options] <filename(s)>
General options:
-h/--help Print this help message and exits.
-v/--version Print the version information and exits.
-q/--quality <INT> Trim low-quality ends from reads in addition to adapter removal.
For RRBS samples, quality trimming will be performed first, and
adapter trimming is carried in a second round. Other files are quality
and adapter trimmed in a single pass. The algorithm is the same as
the one used by BWA (Subtract INT 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). Default Phred
score: 20.
--phred33 Instructs Cutadapt to use ASCII+33 quality scores as Phred scores
(Sanger/Illumina 1.9+ encoding) for quality trimming. Default: ON.
--phred64 Instructs Cutadapt to use ASCII+64 quality scores as Phred scores
(Illumina 1.5 encoding) for quality trimming.
--fastqc Run FastQC in the default mode on the FastQ file once trimming is
complete.
--fastqc_args "<ARGS>" Passes extra arguments to FastQC. If more than one
argument is to be passed to FastQC they must be in the form "arg1
arg2 etc.". An example would be: --fastqc_args "--
nogroup --outdir /home/". Passing extra arguments will
automatically invoke FastQC, so --fastqc does not have to be
specified separately.
-a/--adapter <STRING> Adapter sequence to be trimmed. If not specified explicitly, Trim
Galore will try to auto-detect whether the Illumina universal,
Nextera transposase or Illumina small RNA adapter sequence was
used. Also see '--illumina', '--nextera' and '--
small_rna'. If no adapter can be detected within the first 1
million sequences of the first file specified Trim Galore defaults to
'--illumina'.
-a2/--adapter2 <STRING> Optional adapter sequence to be trimmed off read 2 of paired-
end files. This option requires '--paired' to be specified as well.
--illumina Adapter sequence to be trimmed is the first 13bp of the Illumina
universal adapter 'AGATCGGAAGAGC' instead of the default auto-
detection of adapter sequence.
--nextera Adapter sequence to be trimmed is the first 12bp of the Nextera
adapter 'CTGTCTCTTATA' instead of the default auto-detection of
adapter sequence.
--small_rna Adapter sequence to be trimmed is the first 12bp of the Illumina
Small RNA Adapter 'ATGGAATTCTCG' instead of the default auto-
detection of adapter sequence.
-s/--stringency <INT> Overlap with adapter sequence required to trim a sequence.
Defaults to a very stringent setting of '1', i.e. even a single bp of
overlapping sequence will be trimmed of the 3' end of any read.
-e <ERROR RATE> Maximum allowed error rate (no. of errors divided by the length of
the matching region) (default: 0.1).
--gzip Compress the output file with gzip. If the input files are gzip-
compressed the output files will be automatically gzip
compressed as well.
--dont_gzip Output files won't be compressed with gzip. This overrides --
gzip.
--length <INT> Discard reads that became shorter than length INT because of either
quality or adapter trimming. A value of '0' effectively disables this
behaviour. Default: 20 bp.
For paired-end files, both reads of a read-pair need to be longer
than <INT> bp to be printed out to validated paired-end files (see
option --paired). If only one read became too short there is the
possibility of keeping such unpaired single-end reads
(see --retain_unpaired). Default pair-cutoff: 20 bp.
-o/--output_dir <DIR> If specified all output will be written to this directory instead of the
current directory.
--no_report_file If specified no report file will be generated.
--suppress_warn If specified any output to STDOUT or STDERR will be suppressed.
--clip_R1 <int> Instructs Trim Galore to remove <int> bp from the 5' end of read
1 (or single-end reads). This may be useful if the qualities were very
poor, or if there is some sort of unwanted bias at the 5' end. Default:
OFF.
--clip_R2 <int> Instructs Trim Galore to remove <int> bp from the 5' end of read 2
(paired-end reads only). This may be useful if the qualities were very
poor, or if there is some sort of unwanted bias at the 5' end. For
paired-end BS-Seq, it is recommended to remove the first few bp
because the end-repair reaction may introduce a bias towards low
methylation. Please refer to the M-bias plot section in the Bismark
User Guide for some examples. Default: OFF.
--three_prime_clip_R1 <int> Instructs Trim Galore to remove <int> bp from the 3' end of
read 1 (or single-end reads) AFTER adapter/quality trimming has
been performed. This may remove some unwanted bias from the 3'
end that is not directly related to adapter sequence or basecall
quality. Default: OFF.
--three_prime_clip_R2 <int> Instructs Trim Galore to remove <int> bp from the 3' end of
read 2 AFTER adapter/quality trimming has been performed. This
may remove some unwanted bias from the 3' end that is not directly
related to adapter sequence or basecall quality. Default: OFF.
RRBS-specific options (MspI digested material):
--rrbs Specifies that the input file was an MspI digested RRBS sample
(recognition site: CCGG). Sequences which were adapter-trimmed
will have a further 2 bp removed from their 3' end. This is to avoid
that the filled-in C close to the second MspI site in a sequence is
used for methylation calls. Sequences which were merely trimmed
because of poor quality will not be shortened further.
--non_directional Selecting this option for non-directional RRBS libraries will screen
quality-trimmed sequences for 'CAA' or 'CGA' at the start of the
read and, if found, removes the first two basepairs. Like with the
option '--rrbs' this avoids using cytosine positions that were
filled-in during the end-repair step. '--non_directional'
requires '--rrbs' to be specified as well.
--keep Keep the quality trimmed intermediate file. Default: off, i.e. the
temporary file is being deleted after adapter trimming. Only has an
effect for RRBS samples since other FastQ files are not trimmed for
poor qualities separately.
Note for RRBS using MseI:
If your DNA material was digested with MseI (recognition motif: TTAA) instead of MspI it is NOT
necessary to specify --rrbs or --non_directional since virtually all reads should start with
the sequence 'TAA', and this holds true for both directional and non-directional libraries. As the end-
repair of 'TAA' restricted sites does not involve any cytosines it does not need to be treated
especially. Instead, simply run Trim Galore! in the standard (i.e. non-RRBS) mode.
Paired-end specific options:
--paired This option performs length trimming of quality/adapter/RRBS
trimmed reads for paired-end files. To pass the validation test, both
sequences of a sequence pair are required to have a certain
minimum length which is governed by the option --length (see
above). If only one read passes this length threshold the other read
can be rescued (see option --retain_unpaired). Using this
option lets you discard too short read pairs without disturbing the
sequence-by-sequence order of FastQ files which is required by
many aligners.
Trim Galore! expects paired-end files to be supplied in a
pairwise fashion, e.g. file1_1.fq file1_2.fq
SRR2_1.fq.gz SRR2_2.fq.gz ... .
-t/--trim1 Trims 1 bp off every read from its 3' end. This may be needed for
FastQ files that are to be aligned as paired-end data with Bowtie.
This is because Bowtie (1) regards alignments like this:
R1 --------------------------->
R2 <---------------------------
or this:
R1 ----------------------->
R2 <-----------------
as invalid (whenever a start/end coordinate is contained within the
other read).
--retain_unpaired If only one of the two paired-end reads became too short, the
longer read will be written to either '.unpaired_1.fq' or
'.unpaired_2.fq' output files. The length cutoff for unpaired
single-end reads is governed by the parameters -r1/--length_1
and -r2/--length_2. Default: OFF.
-r1/--length_1 <INT> Unpaired single-end read length cutoff needed for read 1 to be
written to '.unpaired_1.fq' output file. These reads may be
mapped in single-end mode. Default: 35 bp.
-r2/--length_2 <INT> Unpaired single-end read length cutoff needed for read 2 to be
written to '.unpaired_2.fq' output file. These reads may be
mapped in single-end mode. Default: 35 bp.
Changelog:
16-07-14: Version 0.3.7 released
o Applied small change that makes paired-end mode work again (it
was accidentally broken by changing @ARGV for @filenames when
looping through the filenames...)
11-07-14: Version 0.3.6 released
o Added the new options '--three_prime_clip_r1' and '--
three_prime_clip_r2' to clip any number of bases from the 3' end
after adapter/quality trimming has completed
o Added a check to see if Cutadapt exits fine. Else, Trim Galore will
bail a well
o The option '--stringency' needs to be spelled out now since using -s
was ambiguous because of '--suppress_warn'
late 2013: Version 0.3.5 released
o Added the Trim Galore version number to the summary report
19-09-13: Version 0.3.4 released
o Added single-end or paired-end mode to the summary report
o In paired-end mode, the Read 1 summary report will no longer state
that no sequence have been discarded due to trimming. This will be
stated in the trimming report of Read 2 once the validation step has
been completed
10-09-13: Version 0.3.3 released
o Fixed a bug what was accidentally introduced which would add an
additional empty line in single-end trimming mode
03-09-13: Version 0.3.2 released
o Specifying --clip_R1 or --clip_R2 will no longer attempt to clip
sequences that have been adapter- or quality-trimmed below the
clipping threshold
o Specifying an output directory with --rrbs mode should now correctly
create temporary files
15-07-13: Version 0.3.1 released
o The default length cutoff is now set at an earlier timepoint to avoid a
clash in paired-end mode when '--retain_unpaired' and individual
read lengths for read 1 and read 2 had been defined
15-07-13: Version 0.3.0 released
o Added the options '--clip_R1' and '--clip_R2' to trim off a fixed amount
of bases at from the 5' end of reads. This can be useful if the quality
is unusually low at the start, or whenever there is an undesired bias
at the start of reads. An example for this could be PBAT-Seq in
general, or the start of read 2 for every bisulfite-Seq paired-end
library where end repair procedure introduces unmethylated
cytosines. For more information on this see the M-bias section of
the Bismark User Guide
10-04-13: Version 0.2.8 released
o Trim Galore will now compress output files with GZIP on the fly
instead of compressing the trimmed file once trimming has
completed. In the interest of time temporary files are not being
compressed
o Added a small sanity check to exit if no files were supplied for
trimming. Thanks to P. for 'bringing this to my attention'
01-03-13: Version 0.2.7 released
o Added a new option '--dont_gzip' that will force the output files not to
be gzip compressed. This overrides both the '--gzip' option or a .gz
line ending of the input file(s)
07-02-13: Version 0.2.6 released
o Fixes some bugs which would not gzip or run FastQC correctly when
the option '-o' had been specified
o When '--fastqc' is specified in paired-end mode the intermediate files
'_trimmed.fq' are no longer analysed (only files '_val_1' and '_val_2')
19-10-12: Version 0.2.5 released
o Added option '-o/--output_directory' to redirect all output (including
temporary files) to another folder (required for implementation into
Galaxy)
o Added option '--no_report_file' to suppress a report file
o Added option '--suppress_warn' to suppress any output to STDOUT
or STDERR
02-10-12: Version 0.2.4 released
o Removed the shorthand '-l' from the description as it might conflict
with the paired-end options '-r1/--length1' or '-r2/--length2'. Please
use '--length' instead
o Changed the reporting to show the true Phred score quality cutoff
o Corrected a typo in stringency...
31-07-12: Version 0.2.3 released
o Added an option '-e ERROR RATE' that allows one to specify the
maximum error rate for trimming manually (the default is 0.1)
09-05-12: Version 0.2.2 released
o Added an option '-a2/--adapter2' so that one can specify individual
adapter sequences for the two reads of paired-end files; hereby the
sequence provided as '-a/--adapter' is used to trim read 1, and the
sequence provided as '-a2/--adapter2' is used to trim read 2. This
option requires '--paired' to be specified as well
20-04-12: Version 0.2.1 released
o Trim Galore! now has an option '--paired' which has the same
functionality as the validate_paired_ends script we offered
previously. This option discards read pairs if one (or both) reads of a
read pair became shorter than a given length cutoff
o Reads of a read-pair that are longer than a given threshold but for
which the partner read has become too short can optionally be
written out to single-end files. This ensures that the information of a
read pair is not lost entirely if only one read is of good quality
o Paired-end reads may be truncated by a further 1 bp from their 3' end
to avoid problems with invalid alignments with Bowtie 1 (which
regards alignments that contain each other as invalid...)
o The output may be gzip compressed (this happens automatically if
the input files were gzipped (i.e. end in .gz))
o The documentation was extended substantially. We also added
some recommendations for RRBS libraries for MseI digested
material (recognition motif TTAA)
21-03-12: Version 0.1.4 released
o Phred33 (Sanger) encoding is now the default quality scheme
o Fixed a bug for Phred64 encoding that would occur if several files
were specified at once
14-03-12: Version 0.1.3 released
o Initial stand-alone release; all basic functions working
o Added the option 'fastqc_args' to pass extra options to FastQC
