An ultra-fast all-in-one FASTQ preprocessor (QC/adapters/trimming/filtering/splitting/merging...)
A tool designed to provide fast all-in-one preprocessing for FastQ files. This tool is developed in C++ with multithreading supported to afford high performance. - fastp - features - simple usage - examples of report - get fastp - install with Bioconda - or download binary (only for Linux systems, http://opengene.org/fastp/fastp) - or compile from source - compile from source for windows user with MinGW64-distro - input and output - output to STDOUT - input from STDIN - store the unpaired reads for PE data - store the reads that fail the filters - process only part of the data - do not overwrite exiting files - split the output to multiple files for parallel processing - merge PE reads - filtering - quality filter - length filter - low complexity filter - Other filter - adapters - per read cutting by quality score - base correction for PE data - global trimming - polyG tail trimming - polyX tail trimming - unique molecular identifier (UMI) processing - UMI example - output splitting - splitting by limiting file number - splitting by limiting the lines of each file - overrepresented sequence analysis - merge paired-end reads - all options - citation
This tool is being intensively developed, and new features can be implemented soon if they are considered useful. If you have any additional requirement for
fastp, please file an issue:https://github.com/OpenGene/fastp/issues/new
fastp -i in.fq -o out.fq
fastp -i in.R1.fq.gz -I in.R2.fq.gz -o out.R1.fq.gz -O out.R2.fq.gzBy default, the HTML report is saved to
fastp.html(can be specified with
-hoption), and the JSON report is saved to
fastp.json(can be specified with
fastpcreates reports in both HTML and JSON format. * HTML report: http://opengene.org/fastp/fastp.html * JSON report: http://opengene.org/fastp/fastp.json
conda install -c bioconda fastp ```
# this binary was compiled on CentOS, and tested on CentOS/Ubuntu wget http://opengene.org/fastp/fastp chmod a+x ./fastp
# get source (you can also use browser to download from master or releases) git clone https://github.com/OpenGene/fastp.git
cd fastp make
sudo make install
Get compiler from https://nuwen.net/mingw.html and install as
How to installsection.
git clone -b master --depth=1 https://github.com/OpenGene/fastp.git cd fastp make ## add fastp to your PATH
fastponly relies on
zlib, which is already available on most Linux-like systems. If you get an error like
undefined reference to gzbufferwhen compiling
fastp, you may have to update the
fastpsupports both single-end (SE) and paired-end (PE) input/output. * for SE data, you only have to specify read1 input by
--in1, and specify read1 output by
--out1. * for PE data, you should also specify read2 input by
--in2, and specify read2 output by
--out2. * if you don't specify the output file names, no output files will be written, but the QC will still be done for both data before and after filtering. * the output will be gzip-compressed if its file name ends with
fastpsupports streaming the passing-filter reads to STDOUT, so that it can be passed to other compressors like
bzip2, or be passed to aligners like
bowtie2. * specify
--stdoutto enable this mode to stream output to STDOUT * for PE data, the output will be interleaved FASTQ, which means the output will contain records like
record1-R1 -> record1-R2 -> record2-R1 -> record2-R2 -> record3-R1 -> record3-R2 ...
--stdinif you want to read the STDIN for processing.
--interleaved_into indicate that. ## store the unpaired reads for PE data
--unpaired1to store the reads that read1 passes filters but its paired read2 doesn't, as well as
--unpaired2for unpaired read2.
--unpaired2can be the same, so the unpaired read1/read2 will be written to the same single file. ## store the reads that fail the filters
--failed_outto specify the file name to store the failed reads.
failure reasonwill be appended to its read name. For example,
failure reasonwill be
paired_read_is_failing. ## process only part of the data If you don't want to process all the data, you can specify
--reads_to_processto limit the reads to be processed. This is useful if you want to have a fast preview of the data quality, or you want to create a subset of the filtered data. ## do not overwrite exiting files You can enable the option
--dont_overwriteto protect the existing files not to be overwritten by
fastp. In this case,
fastpwill report an error and quit if it finds any of the output files (read1, read2, json report, html report) already exists before. ## split the output to multiple files for parallel processing See output splitting ## merge PE reads See merge paired-end reads
Multiple filters have been implemented.
Quality filtering is enabled by default, but you can disable it by
disable_quality_filtering. Currently it supports filtering by limiting the N base number (
-n, --n_base_limit), and the percentage of unqualified bases.
To filter reads by its percentage of unqualified bases, two options should be provided: *
-q, --qualified_quality_phredthe quality value that a base is qualified. Default 15 means phred quality >=Q15 is qualified. *
-u, --unqualified_percent_limithow many percents of bases are allowed to be unqualified (0~100). Default 40 means 40%
You can also filter reads by its average quality score *
-e, --average_qualif one read's average quality score <avg_qual, then this read/pair is discarded. Default 0 means no requirement (int [=0])
Length filtering is enabled by default, but you can disable it by
--disable_length_filtering. The minimum length requirement is specified with
For some applications like small RNA sequencing, you may want to discard the long reads. You can specify
--length_limitto discard the reads longer than
length_limit. The default value 0 means no limitation.
Low complexity filter is disabled by default, and you can enable it by
--low_complexity_filter. The complexity is defined as the percentage of base that is different from its next base (base[i] != base[i+1]). For example: ```
seq = 'AAAATTTTTTTTTTTTTTTTTTTTTGGGGGGGGGGGGGGGGGGGGGGCCCC' complexity = 3/(51-1) = 6% ``
The threshold for low complexity filter can be specified by-Y
. It's range should be0~100`, and its default value is 30, which means 30% complexity is required.
New filters are being implemented. If you have a new idea or new request, please file an issue.
Adapter trimming is enabled by default, but you can disable it by
--disable_adapter_trimming. Adapter sequences can be automatically detected for both PE/SE data. * For SE data, the adapters are evaluated by analyzing the tails of first ~1M reads. This evaluation may be inacurrate, and you can specify the adapter sequence by
--adapter_sequenceoption. If adapter sequence is specified, the auto detection for SE data will be disabled. * For PE data, the adapters can be detected by per-read overlap analysis, which seeks for the overlap of each pair of reads. This method is robust and fast, so normally you don't have to input the adapter sequence even you know it. But you can still specify the adapter sequences for read1 by
--adapter_sequence, and for read2 by
fastpfails to find an overlap (i.e. due to low quality bases), it will use these sequences to trim adapters for read1 and read2 respectively. * For PE data, the adapter sequence auto-detection is disabled by default since the adapters can be trimmed by overlap analysis. However, you can specify
--detect_adapter_for_peto enable it. * For PE data,
fastpwill run a little slower if you specify the sequence adapters or enable adapter auto-detection, but usually result in a slightly cleaner output, since the overlap analysis may fail due to sequencing errors or adapter dimers. * The most widely used adapter is the Illumina TruSeq adapters. If your data is from the TruSeq library, you can add
--adapter_sequence=AGATCGGAAGAGCACACGTCTGAACTCCAGTCA --adapter_sequence_r2=AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGTto your command lines, or enable auto detection for PE data by specifing
fastpcontains some built-in known adapter sequences for better auto-detection. If you want to make some adapters to be a part of the built-in adapters, please file an issue.
You can also specify
--adapter_fastato give a FASTA file to tell
fastpto trim multiple adapters in this FASTA file. Here is a sample of such adapter FASTA file: ```
Illumina TruSeq Adapter Read 1 AGATCGGAAGAGCACACGTCTGAACTCCAGTCA Illumina TruSeq Adapter Read 2 AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT polyA AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA ```
The adapter sequence in this file should be at least 6bp long, otherwise it will be skipped. And you can give whatever you want to trim, rather than regular sequencing adapters (i.e. polyA).
fastpfirst trims the auto-detected adapter or the adapter sequences given by
--adapter_sequence | --adapter_sequence_r2, then trims the adapters given by
--adapter_fastaone by one.
The sequence distribution of trimmed adapters can be found at the HTML/JSON reports.
fastpsupports per read sliding window cutting by evaluating the mean quality scores in the sliding window. From
fastpsupports 3 different operations, and you enable one or all of them: *
-5, --cut_frontmove a sliding window from front (5') to tail, drop the bases in the window if its mean quality is below cutmeanquality, stop otherwise. Default is disabled. The leading N bases are also trimmed. Use
cut_front_window_sizeto set the widnow size, and
cut_front_mean_qualityto set the mean quality threshold. If the window size is 1, this is similar as the Trimmomatic
-3, --cut_tailmove a sliding window from tail (3') to front, drop the bases in the window if its mean quality is below cutmeanquality, stop otherwise. Default is disabled. The trailing N bases are also trimmed. Use
cut_tail_window_sizeto set the widnow size, and
cut_tail_mean_qualityto set the mean quality threshold. If the window size is 1, this is similar as the Trimmomatic
-r, --cut_rightmove a sliding window from front to tail, if meet one window with mean quality < threshold, drop the bases in the window and the right part, and then stop. Use
cut_right_window_sizeto set the widnow size, and
cut_right_mean_qualityto set the mean quality threshold. This is similar as the Trimmomatic
WARNING: all these three operations will interfere deduplication for SE data, and
--cut_rightmay also interfere deduplication for PE data. The deduplication algorithms rely on the exact matchment of coordination regions of the grouped reads/pairs.
--cut_rightis enabled, then there is no need to enable
--cut_tail, since the former is more aggressive. If
--cut_rightis enabled together with
--cut_frontwill be performed first before
--cut_rightto avoid dropping whole reads due to the low quality starting bases.
Please be noted that
--cut_frontwill interfere deduplication for both PE/SE data, and
--cut_tailwill interfere deduplication for SE data, since the deduplication algorithms rely on the exact matchment of coordination regions of the grouped reads/pairs.
If you don't set window size and mean quality threshold for these function respectively,
fastpwill use the values from
overlap analysisfor PE data, which try to find an overlap of each pair of reads. If an proper overlap is found, it can correct mismatched base pairs in overlapped regions of paired end reads, if one base is with high quality while the other is with ultra low quality. If a base is corrected, the quality of its paired base will be assigned to it so that they will share the same quality.
This function is not enabled by default, specify
--correctionto enable it. This function is based on overlapping detection, which has adjustable parameters
overlap_len_require (default 30),
overlap_diff_limit (default 5)and
overlap_diff_limit_percent (default 20%). Please note that the reads should meet these three conditions simultaneously.
fastpsupports global trimming, which means trim all reads in the front or the tail. This function is useful since sometimes you want to drop some cycles of a sequencing run.
For example, the last cycle of Illumina sequencing is uaually with low quality, and it can be dropped with
-T, --trim_tail2. But if these options are not specified, they will be as same as read1 options, which means
trim_front2 = trim_front1and
trim_tail2 = trim_tail1.
-b, --max_len1for read1, and
-B, --max_len2for read2. If
--max_len1is specified but
--max_len2will be same as
--max_len1. For example, if
--max_len1is specified and read1 is longer than
fastpwill trim read1 at its tail to make it as long as
Please note that the trimming for
--max_lenlimitation will be applied at the last step. Following are fastp's processing steps that may orderly affect the read lengthes:
1, UMI preprocessing (--umi) 2, global trimming at front (--trim_front) 3, global trimming at tail (--trim_tail) 4, quality pruning at 5' (--cut_front) 5, quality pruning by sliding window (--cut_right) 6, quality pruning at 3' (--cut_tail) 7, trim polyG (--trim_poly_g, enabled by default for NovaSeq/NextSeq data) 8, trim adapter by overlap analysis (enabled by default for PE data) 9, trim adapter by adapter sequence (--adapter_sequence, --adapter_sequence_r2. For PE data, this step is skipped if last step succeeded) 10, trim polyX (--trim_poly_x) 11, trim to max length (---max_len)
For Illumina NextSeq/NovaSeq data,
polyGcan happen in read tails since
Gmeans no signal in the Illumina two-color systems.
fastpcan detect the polyG in read tails and trim them. This feature is enabled for NextSeq/NovaSeq data by default, and you can specify
--trim_poly_gto enable it for any data, or specify
--disable_trim_poly_gto disable it. NextSeq/NovaSeq data is detected by the machine ID in the FASTQ records.
A minimum length can be set withfor
fastpto detect polyG. This value is 10 by default.
This feature is similar as polyG tail trimming, but is disabled by default. Use
--trim_poly_xto enable it. A minimum length can be set with for
fastpto detect polyX. This value is 10 by default.
polyG tail trimmingand
polyX tail trimmingare both enabled, fastp will perform
polyG trimmingfirst, then perform
polyX trimming. This setting is useful for trimming the tails having
polyX (i.e. polyA)before
polyGis usually caused by sequencing artifacts, while
polyAcan be commonly found from the tails of mRNA-Seq reads.
UMI is useful for duplication elimination and error correction based on generating consensus of reads originated from a same DNA fragment. It's usually used in deep sequencing applications like ctDNA sequencing. Commonly for Illumina platforms, UMIs can be integrated in two different places:
indexor head of
read. To enable UMI processing, you have to enable
--umioption in the command line, and specify
--umi_locto specify the UMI location, it can be one of: *
index1the first index is used as UMI. If the data is PE, this UMI will be used for both read1/read2. *
index2the second index is used as UMI. PE data only, this UMI will be used for both read1/read2. *
read1the head of read1 is used as UMI. If the data is PE, this UMI will be used for both read1/read2. *
read2the head of read2 is used as UMI. PE data only, this UMI will be used for both read1/read2. *
index1_index2is used as UMI for both read1/read2.
umi1as the head of read1, and
umi2as the head of read2.
umi1_umi2is used as UMI for both read1/read2.
--umi_locis specified with
per_read, the length of UMI should specified with
fastpwill extract the UMIs, and append them to the first part of read names, so the UMIs will also be presented in SAM/BAM records. If the UMI is in the reads, then it will be shifted from read so that the read will become shorter. If the UMI is in the index, it will be kept.
A prefix can be specified with
--umi_prefix. If prefix is specified, an underline will be used to connect it and UMI. For example, UMI=AATTCCGG, prefix=UMI, then the final string presented in the name will be
If the UMI location is read1/read2/perread, fastp can skip some bases after UMI to trim the UMI separator and A/T tailing. Specify `--umiskip` to enable the number of bases to skip. By default it is not enabled.
The original read:
@NS500713:64:HFKJJBGXY:1:11101:1675:1101 1:N:0:TATAGCCT+GACCCCCA AAAAAAAAGCTACTTGGAGTACCAATAATAAAGTGAGCCCACCTTCCTGGTACCCAGACATTTCAGGAGGTCGGGAAA + 6AAAAAEEEEE/E/EA/E/AEA6EE//AEE66/AAE//EEE/E//E/AA/EEE/A/AEE/EEA//EEEEEEEE6EEAAAfter it's processed with command:
fastp -i R1.fq -o out.R1.fq -U --umi_loc=read1 --umi_len=8:
@NS500713:64:HFKJJBGXY:1:11101:1675:1101:AAAAAAAA 1:N:0:TATAGCCT+GACCCCCA GCTACTTGGAGTACCAATAATAAAGTGAGCCCACCTTCCTGGTACCCAGACATTTCAGGAGGTCGGGAAA + EEE/E/EA/E/AEA6EE//AEE66/AAE//EEE/E//E/AA/EEE/A/AEE/EEA//EEEEEEEE6EEAA
For parallel processing of FASTQ files (i.e. alignment in parallel),
fastpsupports splitting the output into multiple files. The splitting can work with two different modes:
by limiting file numberor
by limiting lines of each file. These two modes cannot be enabled together.
The file names of these split files will have a sequential number prefix, adding to the original file name specified by
--out2, and the width of the prefix is controlled by the
--split_prefix_digitsoption. For example,
--split=3, then the output files will be
--splitto specify how many files you want to have.
fastpevaluates the read number of a FASTQ by reading its first ~1M reads. This evaluation is not accurate so the file sizes of the last several files can be a little differnt (a bit bigger or smaller). For best performance, it is suggested to specify the file number to be a multiple of the thread number.
--split_by_linesto limit the lines of each file. The last files may have smaller sizes since usually the input file cannot be perfectly divided. The actual file lines may be a little greater than the value specified by
fastpreads and writes data by blocks (a block = 1000 reads).
Overrepresented sequence analysis is disabled by default, you can specify
--overrepresentation_analysisto enable it. For consideration of speed and memory,
fastponly counts sequences with length of 10bp, 20bp, 40bp, 100bp or (cycles - 2 ).
By default, fastp uses 1/20 reads for sequence counting, and you can change this settings by specifying
--overrepresentation_samplingoption. For example, if you set
-P 100, only 1/100 reads will be used for counting, and if you set
-P 1, all reads will be used but it will be extremely slow. The default value 20 is a balance of speed and accuracy.
fastpnot only gives the counts of overrepresented sequence, but also gives the information that how they distribute over cycles. A figure is provided for each detected overrepresented sequence, from which you can know where this sequence is mostly found.
For paired-end (PE) input, fastp supports stiching them by specifying the
-m/--mergeoption. In this
--merged_outshouuld be given to specify the file to store merged reads, otherwise you should enable
--stdoutto stream the merged reads to STDOUT. The merged reads are also filtered.
--out2will be the reads that cannot be merged successfully, but both pass all the filters.
--unpaired1will be the reads that cannot be merged,
read1passes filters but
--unpaired2will be the reads that cannot be merged,
read2passes filters but
--include_unmergedcan be enabled to make reads of
--merged_out. So you will get a single output file. This option is disabled by default.
--failed_outcan still be given to store the reads (either merged or unmerged) failed to passing filters.
In the output file, a tag like
merged_xxx_yyywill be added to each read name to indicate that how many base pairs are from read1 and from read2, respectively. For example,
@NB551106:9:H5Y5GBGX2:1:22306:18653:13119 1:N:0:GATCAG merged_150_15means that 150bp are from read1, and 15bp are from read2.
fastpprefers the bases in read1 since they usually have higher quality than read2.
Same as the base correction feature, this function is also based on overlapping detection, which has adjustable parameters
overlap_len_require (default 30),
overlap_diff_limit (default 5)and
overlap_diff_limit_percent (default 20%). Please note that the reads should meet these three conditions simultaneously.
usage: fastp -i -o [-I -O ] [options...] options: # I/O options -i, --in1 read1 input file name (string) -o, --out1 read1 output file name (string [=]) -I, --in2 read2 input file name (string [=]) -O, --out2 read2 output file name (string [=]) --unpaired1 for PE input, if read1 passed QC but read2 not, it will be written to unpaired1. Default is to discard it. (string [=]) --unpaired2 for PE input, if read2 passed QC but read1 not, it will be written to unpaired2. If --unpaired2 is same as --unpaired1 (default mode), both unpaired reads will be written to this same file. (string [=]) --failed_out specify the file to store reads that cannot pass the filters. (string [=]) --overlapped_out for each read pair, output the overlapped region if it has no any mismatched base. (string [=]) -m, --merge for paired-end input, merge each pair of reads into a single read if they are overlapped. The merged reads will be written to the file given by --merged_out, the unmerged reads will be written to the files specified by --out1 and --out2. The merging mode is disabled by default. --merged_out in the merging mode, specify the file name to store merged output, or specify --stdout to stream the merged output (string [=]) --include_unmerged in the merging mode, write the unmerged or unpaired reads to the file specified by --merge. Disabled by default. -6, --phred64 indicate the input is using phred64 scoring (it'll be converted to phred33, so the output will still be phred33) -z, --compression compression level for gzip output (1 ~ 9). 1 is fastest, 9 is smallest, default is 4. (int [=4]) --stdin input from STDIN. If the STDIN is interleaved paired-end FASTQ, please also add --interleaved_in. --stdout output passing-filters reads to STDOUT. This option will result in interleaved FASTQ output for paired-end input. Disabled by default. --interleaved_in indicate that is an interleaved FASTQ which contains both read1 and read2. Disabled by default. --reads_to_process specify how many reads/pairs to be processed. Default 0 means process all reads. (int [=0]) --dont_overwrite don't overwrite existing files. Overwritting is allowed by default. --fix_mgi_id the MGI FASTQ ID format is not compatible with many BAM operation tools, enable this option to fix it.
adapter trimming options
-A, --disable_adapter_trimming adapter trimming is enabled by default. If this option is specified, adapter trimming is disabled -a, --adapter_sequence the adapter for read1. For SE data, if not specified, the adapter will be auto-detected. For PE data, this is used if R1/R2 are found not overlapped. (string [=auto]) --adapter_sequence_r2 the adapter for read2 (PE data only). This is used if R1/R2 are found not overlapped. If not specified, it will be the same as (string [=]) --adapter_fasta specify a FASTA file to trim both read1 and read2 (if PE) by all the sequences in this FASTA file (string [=]) --detect_adapter_for_pe by default, the adapter sequence auto-detection is enabled for SE data only, turn on this option to enable it for PE data.
global trimming options
-f, --trim_front1 trimming how many bases in front for read1, default is 0 (int [=0]) -t, --trim_tail1 trimming how many bases in tail for read1, default is 0 (int [=0]) -b, --max_len1 if read1 is longer than max_len1, then trim read1 at its tail to make it as long as max_len1. Default 0 means no limitation (int [=0]) -F, --trim_front2 trimming how many bases in front for read2. If it's not specified, it will follow read1's settings (int [=0]) -T, --trim_tail2 trimming how many bases in tail for read2. If it's not specified, it will follow read1's settings (int [=0]) -B, --max_len2 if read2 is longer than max_len2, then trim read2 at its tail to make it as long as max_len2. Default 0 means no limitation. If it's not specified, it will follow read1's settings (int [=0])
polyG tail trimming, useful for NextSeq/NovaSeq data
-g, --trim_poly_g force polyG tail trimming, by default trimming is automatically enabled for Illumina NextSeq/NovaSeq data --poly_g_min_len the minimum length to detect polyG in the read tail. 10 by default. (int [=10]) -G, --disable_trim_poly_g disable polyG tail trimming, by default trimming is automatically enabled for Illumina NextSeq/NovaSeq data
polyX tail trimming
-x, --trim_poly_x enable polyX trimming in 3' ends. --poly_x_min_len the minimum length to detect polyX in the read tail. 10 by default. (int [=10])
per read cutting by quality options
-5, --cut_front move a sliding window from front (5') to tail, drop the bases in the window if its mean quality < threshold, stop otherwise. -3, --cut_tail move a sliding window from tail (3') to front, drop the bases in the window if its mean quality < threshold, stop otherwise. -r, --cut_right move a sliding window from front to tail, if meet one window with mean quality < threshold, drop the bases in the window and the right part, and then stop. -W, --cut_window_size the window size option shared by cut_front, cut_tail or cut_sliding. Range: 1
1000, default: 4 (int [=4]) -M, --cut_mean_quality the mean quality requirement option shared by cut_front, cut_tail or cut_sliding. Range: 136 default: 20 (Q20) (int [=20]) --cut_front_window_size the window size option of cut_front, default to cut_window_size if not specified (int [=4]) --cut_front_mean_quality the mean quality requirement option for cut_front, default to cut_mean_quality if not specified (int [=20]) --cut_tail_window_size the window size option of cut_tail, default to cut_window_size if not specified (int [=4]) --cut_tail_mean_quality the mean quality requirement option for cut_tail, default to cut_mean_quality if not specified (int [=20]) --cut_right_window_size the window size option of cut_right, default to cut_window_size if not specified (int [=4]) --cut_right_mean_quality the mean quality requirement option for cut_right, default to cut_mean_quality if not specified (int [=20])
quality filtering options
-Q, --disable_quality_filtering quality filtering is enabled by default. If this option is specified, quality filtering is disabled -q, --qualified_quality_phred the quality value that a base is qualified. Default 15 means phred quality >=Q15 is qualified. (int [=15]) -u, --unqualified_percent_limit how many percents of bases are allowed to be unqualified (0
100). Default 40 means 40% (int [=40]) -n, --n_base_limit if one read's number of N base is >n_base_limit, then this read/pair is discarded. Default is 5 (int [=5]) -e, --average_qual if one read's average quality score =1000), a sequential number prefix will be added to output name ( 0001.out.fq, 0002.out.fq...), disabled by default (long [=0]) -d, --split_prefix_digits the digits for the sequential number padding (110), default is 4, so the filename will be padded as 0001.xxx, 0 to disable padding (int [=4])
-?, --help print this message
Shifu Chen, Yanqing Zhou, Yaru Chen, Jia Gu; fastp: an ultra-fast all-in-one FASTQ preprocessor, Bioinformatics, Volume 34, Issue 17, 1 September 2018, Pages i884–i890, https://doi.org/10.1093/bioinformatics/bty560