Development manual for vsearch 2.32.0, built from the dev branch: it describes changes that are not released yet. The manual for the current release is published separately.

NAME

vsearch --usearch_global — search sequences against a reference database

SYNOPSIS

DESCRIPTION

The vsearch command --usearch_global searches the query sequences in a fasta or fastq file against a reference database (--db), using global pairwise alignment (Needleman-Wunsch). The database can be a fasta or fastq file, or a preformatted UDB database (see vsearch-makeudb_usearch(1)).

For each query, vsearch first pre-filters the database by counting shared k-mers (words), then performs global pairwise alignments on the most promising candidates. By default, the search stops after --maxaccepts hits are accepted or --maxrejects candidates fail the identity threshold. Setting both to 0 considers every candidate the pre-filter kept; add --minwordmatches 0 to drop the pre-filter as well and compare the query against the whole database. Using values of --id below 0.5 is unlikely to capture additional hits due to the k-mer pre-filter.

Alignment is global, not local: a query is aligned end-to-end against a target rather than searched for as a subsequence, and at most one alignment is reported per database sequence and per strand. A motif occurring several times within one long target therefore yields a single hit, and two at most with --strand both. Enumerating every occurrence calls for a local-alignment tool instead.

The identity threshold is set with --id. By default, only the plus strand of the query is compared to the database; use --strand both to also check the reverse complement. Masking is applied with --qmask (queries) and --dbmask (database).

At least one output option must be specified. This command is multi-threaded: the queries are distributed over the available threads, so the order of the entries written to --alnout, --blast6out, --fastapairs, --lcaout, --matched, --notmatched, --qsegout, --samout, --tsegout, --uc and --userout may vary from run to run when more than one thread is used. The --biomout, --dbmatched, --dbnotmatched, --mothur_shared_out and --otutabout tables are assembled after the search, or written in database order, and keep a stable order. The results themselves do not depend on the thread count.

To illustrate a search at 97% identity:

Query file:    Database:       Results (--blast6out):

>q1            >t1             q1  t1  97.5  ...
ACGTACGT  -->  ACGTAGGT  -->   q2  *   *     ... (no hit; this line is
>q2            >t2                 only written with --output_no_hits)
TTTTTTTT       ACGTACGT

OPTIONS

mandatory options

--db and --id must both be specified, along with at least one output option.

--db filename

Compare query sequences against the target sequences in filename, in fasta or fastq format. Alternatively, the name of a preformatted UDB database created with --makeudb_usearch may be specified (see vsearch-makeudb_usearch(1) and vsearch-udb(5)). --db accepts - to read the database from standard input, as well as an explicit stream path such as /dev/stdin, a named pipe, or a process substitution. The query and the database cannot both be -, however, as they would compete for the same standard input; give at least one of them an explicit path.

--id real

Reject the match if the pairwise identity with the target sequence is lower than real (value ranging from 0.0 to 1.0 included). The pairwise identity is defined by default as (matching columns) / (alignment length - terminal gaps). That definition can be modified with --iddef.

A column holding an ambiguous symbol counts as a matching column whenever the two symbols share at least one of the nucleotides they represent, so an N matches anything: a query aligned over a run of Ns is reported at 100% identity. Use --n_mismatch to count these columns as mismatches instead.

Which pairs reach the alignment stage where --id is applied is decided beforehand by a k-mer pre-filter (see --minwordmatches and --wordlength): a pair sharing too few words is never aligned and never reported, whatever its identity. Short or heavily masked sequences share few words, so lowering --id alone does not make them match; lower --minwordmatches too. Below an --id of about 0.5, it is the pre-filter rather than --id that decides the outcome.

core options

--dbmask none|dust|soft: Mask regions in the database sequences using the dust method or the soft method, or none to suppress masking. See vsearch-fastx_mask(1) for more details. Warning, when using soft masking, search commands become case sensitive: masking excludes masked regions from the k-mer index used to select candidate targets (the pairwise alignment itself always ignores case). A database sequence with no unmasked stretch of at least the word length contributes no k-mer to the index and is never selected as a candidate target; it can still be reported when the query itself samples no k-mer, as such a query is compared against every database sequence (see --qmask). vsearch warns when there are such sequences, giving how many. The default is to mask using dust.
--iddef 0|1|2|3|4: Change the pairwise identity definition used with --id. Accepted values are:

CD-HIT definition: (matching columns) / (shortest sequence length).
edit distance: (matching columns) / (alignment length).
edit distance excluding terminal gaps (default definition for --id).
Marine Biological Lab definition, counting each gap opening (internal or terminal) as a single mismatch, whether or not the gap was extended: 1.0 - [(mismatches + gap openings)/(longest sequence length)].
BLAST definition, equivalent to --iddef 1 for global pairwise alignments.

In all five definitions, a column holding an ambiguous symbol is a matching column whenever the two symbols share at least one of the nucleotides they represent; an N is thus a match against anything, unless --n_mismatch is given.

All five count columns of the chosen alignment; none of them reads the alignment score. The scoring options (--match, --mismatch, --gapopen, --gapext) therefore act on identity only indirectly, by changing which alignment is optimal, and a pair whose optimal alignment does not change keeps the identity it had. See vsearch-pairwise_alignment_parameters(7).

--maxaccepts positive integer

Set the maximum number of matching target sequences to accept before stopping the search for a given query. The default value is 1. Use together with --maxrejects. If both --maxaccepts and --maxrejects are set to 0, the complete database is searched, save for the targets the word pre-filter removes beforehand (see --minwordmatches).

Target sequences are considered in order of decreasing number of words shared with the query, a proxy for similarity, and each is aligned and then accepted or rejected according to --id and the other criteria. Raising --maxaccepts does not by itself place a query in several clusters or report several hits: it widens the set the outcome is chosen from. When clustering, the query joins the accepted centroid with the highest identity, or the most abundant one with --sizeorder. When searching, --maxhits and --top_hits_only decide how many accepted targets are reported.

--maxrejects positive integer

Set the maximum number of non-matching target sequences to consider before stopping the search for a given query. The default value is 32. Use together with --maxaccepts. If both --maxaccepts and --maxrejects are set to 0, the complete database is searched, save for the targets the word pre-filter removes beforehand (see --minwordmatches).

--qmask none|dust|soft

Mask regions in query sequences using the dust method or the soft method, or none to suppress masking. Values are case-insensitive, so DUST, Dust, and dust are all accepted. See vsearch-fastx_mask(1) for more details. Warning, when using soft masking, search commands become case sensitive: masking excludes masked regions from the k-mer pre-filter that selects candidate targets (the pairwise alignment itself always ignores case). A query with no unmasked stretch of at least the word length samples no k-mers and is therefore compared against every database sequence. The default is to mask using dust.

--strand plus|both

Check the plus strand only (default), or check both strands when comparing sequences. Keywords are case-insensitive.

--threads positive integer

Set the number of computation threads to use, from 1 to 1024. The number of threads should not exceed the number of available CPU cores. The value 0 is also accepted and, like the default, uses all available cores; decimal values are truncated to their integer part. On Linux, “available” accounts for the CPU affinity mask and the cgroup CPU quota of the running process, so a job confined by taskset(1), Slurm, Docker or Kubernetes launches one thread per core it was actually granted, rather than one per core the machine has.

secondary options

--alnout filename

Write pairwise global alignments to filename in a human-readable format. Use --rowlen to set the alignment line width.

--biomout filename

Write an OTU table to filename in the biom version 1.0 JSON file format. The OTUs are represented by the cluster centroids. Sample identifiers are extracted from sequence headers (;sample=abc123; or ;barcodelabel=abc123; patterns, or the initial part of the header). OTU identifiers are extracted from centroid headers (;otu=def789; pattern, or the initial part of the header, or via relabelling options). Taxonomy information is extracted from centroid headers (;tax=...; pattern) if available. Abundance annotations (;size=integer) present in sequence headers are always used when filling the table, whether or not --sizein is given (unlike the --uc cluster summaries, which count each sequence as 1 without --sizein). OTU identifiers must be unique: two OTUs sharing the same identifier are reported as a single observation, and their abundances are summed. When clustering, a relabelling option (--relabel, --relabel_self, --relabel_md5 or --relabel_sha1) guarantees unique identifiers; when searching, the database itself must have unique headers.

--blast6out filename

Write results to filename using a BLAST-like tab-separated format with twelve fields per query-target match: query label, target label, percentage identity, alignment length, mismatches, gap openings, query start, query end, target start, target end, expectation value (always -1), and bit score (always 0). If --output_no_hits is used, non-matching queries are also written. Note that vsearch uses global pairwise alignments, not BLAST’s seed-and-extend algorithm.

--bzip2_decompress

Specify that the input pipe is streaming data compressed using Huffman coding. See bzip2(1) for more details. This option is required when compressed data arrives on standard input through a pipe (‘-’), where the format cannot be detected without consuming the stream. It is not needed when reading from a regular file compressed with bzip2, nor when such a file is redirected to standard input: compression is then detected automatically, and a contradicting option is ignored (with a warning when the input is standard input). Pipes other than standard input, such as shell process substitutions and named FIFOs, are always read as uncompressed data; compressed data must arrive on standard input or as a named file.

--dbmatched filename

Write database target sequences that match at least one query sequence to filename, in fasta format. If --sizeout is specified, the number of matching queries is annotated in the header (;size=integer).

--dbnotmatched filename

Write database target sequences that do not match any query sequence to filename, in fasta format.

--fasta_width positive integer

Set the maximal width of sequences when writing fasta files. Longer sequences are folded and written on several lines. Default width is 80 nucleotides. Set to zero (0) to suppress folding.

--fastapairs filename

Write pairwise alignments of query and target sequences to filename, in fasta format.

--gzip_decompress

Specify that the input pipe is streaming data compressed using Lempel-Ziv coding. See gzip(1) for more details. This option is required when compressed data arrives on standard input through a pipe (‘-’), where the format cannot be detected without consuming the stream. It is not needed when reading from a regular file compressed with gzip, nor when such a file is redirected to standard input: compression is then detected automatically, and a contradicting option is ignored (with a warning when the input is standard input). Pipes other than standard input, such as shell process substitutions and named FIFOs, are always read as uncompressed data; compressed data must arrive on standard input or as a named file.

--hardmask

Replace masked nucleotides with Ns, rather than lowercasing them.

This is also the only masking option that reaches the alignment. Soft and dust masking (see --qmask and --dbmask) only keep masked words out of the k-mer pre-filter that selects candidates; the masked region is still aligned and scored like any other. An N, on the other hand, scores zero and counts as a matching column, so hard masking makes a masked region match whatever it is aligned against, and the reported identity can only go up. Add --n_mismatch to count those columns as mismatches instead.

--idprefix positive integer

Reject the sequence match if the first integer nucleotides of the target do not match the query.

--idsuffix positive integer

Reject the sequence match if the last integer nucleotides of the target do not match the query.

--label_suffix string

Add the suffix string to sequence headers when writing fasta or fastq files. For example, with --label_suffix ";status=healthy", sequence header ‘>seq1’ becomes ‘>seq1;status=healthy’.

--lca_cutoff real

Set the fraction of hits that must agree at a given taxonomic rank for that rank to be included in the last common ancestor (LCA) output (see --lcaout). The default value is 1.0, requiring all hits to agree. Lower values (e.g., 0.95) tolerate a small fraction of disagreeing hits. The value must be greater than 0.5 and at most 1.0.

--lcaout filename

Write last common ancestor (LCA) information to filename in a tab-separated format. The first column contains the query identifier and the second column contains the deepest taxonomic lineage shared by a sufficient fraction of hits (see --lca_cutoff). Database sequence headers must carry taxonomic annotations in the format used by --sintax (e.g., tax=k:Archaea,p:Euryarchaeota,c:Halobacteria). Set --maxaccepts to a value greater than 1 for the LCA to be meaningful. The vote runs over the reported hits (after --maxhits/--top_hits_only, including weak hits from --weak_id). A line is written for every query: queries without hits produce a line with an empty second column, whether or not --output_no_hits is given.

--leftjust

Reject the sequence match if the pairwise alignment begins with gaps.

--lengthout

Add a sequence length annotation (;length=integer) to each sequence header when writing fasta or fastq files.

--log filename

Write messages to filename. Messages include program version, start and finish times, elapsed time, amount of memory available, maximum amount of memory consumed, number of cores and command line options, and if need be, command-specific informational messages, warnings, and errors. Messages are written in addition to the standard error, not instead of it; use --quiet to silence the standard error copy (see the --quiet entry for what it spares).

--matched filename

Write query sequences matching a target sequence to filename, in fasta format.

--maxdiffs positive integer

Reject the sequence match if the alignment contains more than integer substitutions, insertions, or deletions. Terminal gaps are not counted, so a query is never rejected for being shorter than its target; only internal differences count. The diffs userfield reports the quantity compared against.

--maxgaps positive integer

Reject the sequence match if the alignment contains more than integer gap openings: a run of consecutive gap columns counts as one, whatever its length (use --maxdiffs to bound the number of gap columns). Terminal gaps are not counted here either, so a length difference between the query and the target costs no gap opening.

--maxhits non-negative integer

Set the maximum number of hits to report once the search is terminated for a given query; hits are sorted by decreasing identity. Unlimited by default, or when the argument is zero. When searching both strands, --maxhits controls the total number of hits reported per query across both strands.

--maxid real

Reject the sequence match if the pairwise identity between the two sequences is greater than real, a fraction from 0.0 to 1.0 (for example, --maxid 0.97 rejects matches above 97% identity).

--maxqsize positive integer

Reject query sequences with an abundance greater than integer.

--maxqt real

Reject the sequence match if the query/target sequence length ratio is greater than real.

--maxseqlength positive integer

Discard sequences longer than positive integer (50,000 nucleotides by default). The value must not exceed 2,147,481,646 (INT_MAX minus 2,001). When that value is smaller than the effective --minseqlength value (whose default is command-specific), no sequence can pass the length filter: vsearch issues a warning.

--maxsizeratio real

Reject the sequence match if the query/target abundance ratio is greater than real.

--maxsl real

Reject the sequence match if the shorter/longer sequence length ratio is greater than real.

--maxsubs positive integer

Reject the sequence match if the pairwise alignment contains more than integer substitutions.

--mid real

Reject the sequence match if the pairwise identity, computed ignoring all gaps (internal and terminal), is lower than real, a percentage from 0 to 100 (for example, --mid 97 requires 97% identity; note that the other identity options use fractions).

--mincols positive integer

Reject the sequence match if the alignment length is shorter than integer columns.

--minqt real

Reject the sequence match if the query/target sequence length ratio is lower than real.

--minseqlength non-negative integer

Discard sequences shorter than non-negative integer (32 nucleotides by default). A value of 0 retains empty sequences. When that value is larger than the --maxseqlength value, no sequence can pass the length filter: vsearch issues a warning.

--minsizeratio real

Reject the sequence match if the query/target abundance ratio is lower than real.

--minsl real

Reject the sequence match if the shorter/longer sequence length ratio is lower than real.

--mintsize positive integer

Reject target sequences with an abundance lower than integer.

--minwordmatches non-negative integer

Set the minimum number of shared words (i.e. k-mers) required for a target sequence to be considered further. The default value is 12 for the default word length of 8 (see --wordlength); for word lengths 3 to 15 the default values are 18, 17, 16, 15, 14, 12, 11, 10, 9, 8, 7, 5, and 3, respectively. Neither sequence can share more words than it contains, so the requirement is capped by the number of distinct words of the query and by that of the target: when either contains fewer distinct words than the value above, all the words of that sequence must match. If the argument is 0, no word match is required and every target sequence is compared to the query.

Short or heavily masked sequences yield few distinct words, so they may share fewer words than required even when their pairwise identity is high, and such a match is then never reported. When searching or clustering short sequences, lower --minwordmatches (1 is usually enough, and is as sensitive as 0 while much faster), or lower --wordlength.

This word requirement is the only heuristic left once --maxaccepts and --maxrejects are both zero, so --minwordmatches 0 together with them aligns every query against every target: an exhaustive search, at the cost of the speed the pre-filter buys. Older usearch versions spelled --minwordmatches 0 as --nowordcountreject.

--mothur_shared_out filename

Write an OTU table to filename in the mothur ‘shared’ tab-separated plain text format. The first line starts with label, group and numOtus, followed by all OTU identifiers. Each subsequent line starts with vsearch, the sample identifier, the total number of OTUs, and the abundance of each OTU in that sample. Sample and OTU identifiers are extracted from FASTA headers. OTUs are represented by the cluster centroids. Abundance annotations (;size=integer) present in sequence headers are always used when filling the table, whether or not --sizein is given (unlike the --uc cluster summaries, which count each sequence as 1 without --sizein). OTU identifiers must be unique: two OTUs sharing the same identifier are reported in a single column, numOtus counts that column only once, and their abundances are summed. When clustering, a relabelling option (--relabel, --relabel_self, --relabel_md5 or --relabel_sha1) guarantees unique identifiers; when searching, the database itself must have unique headers.

--n_mismatch

Count alignments of nucleotides against Ns as mismatches. By default, an alignment column holding an N is neutral: it scores zero, and it is counted as a matching column when the identity percentage is computed. With --n_mismatch, any column where at least one of the two symbols is an N (regardless of case) is scored and counted as a mismatch instead, N against N included. Both the alignment score and the post-alignment count of matches and mismatches are affected, so the identity percentage compared against --id (and reported in the output files) changes too.

The option targets N only: the other ambiguous symbols (BDHKMRSVWY) keep their default behaviour, and still count as matching any symbol they share a nucleotide with. See vsearch-pairwise_alignment_parameters(7) for the default treatment of ambiguous symbols, and for why a query can align to a long run of Ns with 100% identity.

--no_progress

Suppress the gradually increasing progress indicator normally written to the standard error stderr(3).

--notmatched filename

Write the sequences that were not extracted to filename, in fasta format.

--notrunclabels

Retain whole sequence headers in output files. By default, vsearch truncates sequence headers at first space or tabulation. This option suppresses truncation.

--otutabout filename

Write an OTU table to filename in a classic tab-separated plain text format. The first line starts with #OTU ID followed by sample identifiers. Each subsequent line starts with the OTU identifier followed by the abundances in each sample. Sample and OTU identifiers are extracted from FASTA headers (see --sample). OTUs are represented by the cluster centroids. A taxonomy column is appended if taxonomy information is available for any OTU. Abundance annotations (;size=integer) present in sequence headers are always used when filling the table, whether or not --sizein is given (unlike the --uc cluster summaries, which count each sequence as 1 without --sizein). Rows and columns are both ordered lexicographically by identifier, whatever order the OTUs were created in or the samples first appeared in, and no option changes that. Sorting the table by decreasing abundance is a job for a downstream tool.

OTU identifiers must be unique: two OTUs sharing the same identifier are reported on a single line, and their abundances are summed. When clustering, a relabelling option (--relabel, --relabel_self, --relabel_md5 or --relabel_sha1) guarantees unique identifiers; when searching, the database itself must have unique headers.

--output_no_hits

Write both matching and non-matching queries to --alnout, --blast6out, --samout, or --userout output files. Non-matching queries are labelled ‘No hits’ in --alnout files.

--qsegout filename

Write the aligned part of each query sequence to filename, in fasta format.

--query_cov real

Reject the sequence match if the fraction of the query aligned to the target is lower than real (value ranging from 0.0 to 1.0 included). Query coverage is computed as (matches + mismatches) / query sequence length, not counting internal or terminal gaps. Older usearch versions called this option --queryalnfract.

--quiet

Suppress messages to the standard output stdout(3) and standard error stderr(3), except for warnings and error messages.

--relabel string

Replace sequence headers with the prefix string and a ticker (1, 2, 3, etc.). For example, with --relabel "cluster:", the first sequence header becomes ‘>cluster:1’, the second sequence header becomes ‘>cluster:2’, and so on. To retain annotations, use their corresponding options (--lengthout, --eeout, and --sizeout). Use --relabel_keep to also retain old sequence identifiers.

--relabel_keep

Retain old sequence identifiers by including them at the end of the new headers, after a space.

--relabel_md5

Replace each sequence header with the MD5 digest derived from the sequence itself. The sequence is converted to upper case, and each ‘U’ is replaced with a ‘T’ before computation of the digest. The MD5 digest is a 128-bit value (16 bytes), represented using a string of 32 ASCII characters. Each pair of characters encodes an hexadecimal value, ranging from x00 to xff. See md5(3) for more details, and --relabel_sha1 for an alternative hashing algorithm. To retain annotations, use their corresponding options (--lengthout, --eeout, and --sizeout). Use --relabel_keep to also retain old sequence identifiers.

--relabel_self

Replace each sequence header with the sequence itself. To retain annotations, use their corresponding options (--lengthout, --eeout, and --sizeout). Use --relabel_keep to also retain old sequence identifiers.

--relabel_sha1

Replace each sequence header with the SHA1 digest derived from the sequence itself. The sequence is converted to upper case, and each ‘U’ is replaced with a ‘T’ before computation of the digest. The SHA1 digest is a 160-bit value (20 bytes), represented using a string of 40 ASCII characters. Each pair of characters encodes an hexadecimal value, ranging from x00 to xff. See sha1(3) for more details, and --relabel_md5 for an alternative hashing algorithm. To retain annotations, use their corresponding options (--lengthout, --eeout, and --sizeout). Use --relabel_keep to also retain old sequence identifiers.

--rightjust

Reject the sequence match if the pairwise alignment ends with gaps.

--rowlen positive integer

Set the width of alignment lines in --alnout output. The default value is 64. Set to 0 to disable line wrapping.

--samheader

Include header lines (@HD, @SQ, @PG) in the SAM file produced by --samout. By default, no header lines are written.

--samout filename

Write alignment results to filename in the SAM format, see vsearch-sam(5). Use --samheader to include header lines. Each non-header line is a SAM record representing either a query-target alignment or the absence of a match. The alignment column of each record uses the CIGAR format, see vsearch-cigar(5).

--sample string

Add the given sample identifier string to sequence headers when writing fasta or fastq files. For instance, if string is ‘ABC’, the text ;sample=ABC will be added to the headers. string is silently truncated at the first ‘;’ or whitespace character (space, tab, newline, carriage return, vertical tab or form feed), so such characters should not be used in string. Other characters (alphabetical, numerical and punctuations) are accepted. When nothing is left after truncation — an empty string, or one starting with ‘;’ or a blank character — vsearch issues a warning and writes a bare ;sample= annotation.

--self

Reject the sequence match if the query and target sequence labels are identical. The label is the header up to the first blank, so two records sharing an identifier but carrying different descriptions still reject each other; with --notrunclabels the whole header is the label and they no longer do. Use --selfid to reject on identical sequences rather than identical labels.

--selfid

Reject the sequence match if the query and target sequences are strictly identical.

--sizein

Use the abundance annotations present in sequence headers when reading fasta or fastq file. Search for the pattern [>@;]size=integer[;]. Entries without abundance annotations are silently assumed to be of size=1.

--sizeout

Add abundance annotations to sequence headers when writing fasta or fastq files. Add the pattern ;size=integer. Existing ;size= annotations are reported unchanged; entries without one receive ;size=1. For this command --sizein is not needed: abundance annotations are always parsed from the input headers.

--target_cov real

Reject the sequence match if the fraction of the target sequence aligned to the query is lower than real. Target coverage is computed as (matches + mismatches) / target sequence length, not counting internal or terminal gaps. Older usearch versions called this option --targetalnfract.

--top_hits_only

Report only the hits with the highest pairwise identity for each query.

--tsegout filename

Write the aligned part of each target sequence to filename, in fasta format.

--uc filename

Write search results to filename in a tab-separated uclust-like format with 10 columns. Each query produces a hit (H) or no-hit (N) record. Use --uc_allhits to report all hits per query instead of just the top hit. Columns are:

record type: H (hit) or N (no hit);
ordinal number of the target sequence (zero-based; * for N);
length of the query sequence (* for N);
percentage of identity with the target (* for N);
match orientation + or - (. for N);
not used; 0 (H) or * (N);
not used; 0 (H) or * (N);
CIGAR alignment string (* for N; = if the query and target sequences are strictly identical); see vsearch-cigar(5);
query label;
target label (* for N).

--uc_allhits

When using --uc, report all hits for each query, not just the top hit.

--userfields string

Select and order the fields written to --userout output. Fields are separated by + (e.g. query+target+id). See vsearch-userfields(7) for a complete description of all available fields.

--userout filename

Write user-defined tab-separated output to filename. Select and order the fields with --userfields.

--weak_id real

Report hits with a pairwise identity of at least real, without stopping the search. Unlike --id, weak hits do not count toward --maxaccepts but do count toward --maxrejects. Values larger than the value specified with --id are silently reduced to it. With --cluster_unoise, which does not use --id, real is the identity floor of the denoising step and defaults to 0.90.

--wordlength positive integer

Set the length of words (i.e. k-mers) used for sequence indexing and comparisons. Valid values range from 3 to 15. The default is 8. Note that the default --minwordmatches is derived from this value, so changing one changes both (see --minwordmatches).

Longer words make the k-mer index more selective, so fewer targets are offered as candidates and the search itself gets faster: from word length 5 to 11 the search phase shrank by a factor of 2.4 on 130-nucleotide amplicons and 5.8 on full-length reference sequences. Working against that, the index has 4^wordlength slots, so the memory it needs and the time spent building it both quadruple with each added nucleotide. On a 400 000-sequence database the whole run needed 0.3 GB at word length 10, 1.3 GB at 12 and 16 GB at 15.

The best setting balances the two, and depends on how many queries are searched against a given database: with few queries the index build dominates and a shorter word is cheaper overall, while with many queries the search dominates and a longer word repays its index. Changing the word length is not output-neutral, so it should be chosen for a workload rather than tuned per run.

--xee

Strip expected error (ee) annotations from sequence headers when writing fasta or fastq files. Search for the pattern [>@;]ee=float[;]. Expected error annotations are added by the synonymous options --fastq_eeout and --eeout described in vsearch-fastx_filter(1).

--xlength

Strip sequence length annotations from sequence headers when writing fasta or fastq files. Search for the pattern [>@;]length=integer[;]. Sequence length annotations are added by the --lengthout option.

--xsize

Strip abundance annotations from sequence headers when writing fasta or fastq files. Search for the pattern [>@;]size=integer[;]. Abundance annotations are added by the --sizeout option.

pairwise alignment options

These options modify the pairwise alignment scoring model. Modify with caution.

--gapext string: Set penalties for a gap extension. See vsearch-pairwise_alignment_parameters(7) for a complete description of the gap penalty declaration system. By default, the penalty is set to 2 for extending internal gaps and to 1 for extending terminal gaps, in both query and target sequences.
--gapopen string: Set penalties for a gap opening. See vsearch-pairwise_alignment_parameters(7) for a complete description of the gap penalty declaration system. By default, the penalty is set to 20 for opening internal gaps and to 2 for opening terminal gaps, in both query and target sequences.
--match integer: Set the score assigned to a match (i.e. equivalent nucleotides) in pairwise alignments. The default value is 2. Accepted values range from -32767 to +32767.
--mismatch integer: Set the score assigned to a mismatch (i.e. different nucleotides) in pairwise alignments. The default value is -4. Accepted values range from -32767 to +32767.

ignored options

These options are accepted for compatibility with usearch but have no effect.

--band positive integer: This option is ignored. It is provided for compatibility with usearch.
--fulldp: This option is ignored. It is provided for compatibility with usearch. vsearch always uses a full dynamic programming algorithm (Needleman-Wunsch).
--hspw positive integer: This option is ignored. It is provided for compatibility with usearch.
--minhsp positive integer: This option is ignored. It is provided for compatibility with usearch.
--pattern string: This option is ignored. It is provided for compatibility with usearch.
--slots positive integer: This option is ignored. It is provided for compatibility with usearch.
--xdrop_nw positive integer: This option is ignored. It is provided for compatibility with usearch.

EXAMPLES

Search query sequences against a reference database at 97% identity and write the results in BLAST-like tabular format:

vsearch \
    --usearch_global queries.fasta \
    --db reference.fasta \
    --id 0.97 \
    --blast6out results.tsv

Search both strands, report all hits per query (up to --maxaccepts), and write matching query sequences to a fasta file:

vsearch \
    --usearch_global queries.fasta \
    --db reference.fasta \
    --id 0.97 \
    --strand both \
    --maxaccepts 5 \
    --matched matched.fasta \
    --notmatched unmatched.fasta

Classify query sequences against a taxonomically annotated database and write last common ancestor assignments:

vsearch \
    --usearch_global queries.fasta \
    --db silva_tax.fasta \
    --id 0.97 \
    --maxaccepts 10 \
    --top_hits_only \
    --lcaout taxonomy.tsv \
    --alnout alignments.txt

Build an OTU table from multiple samples (sequences labelled with ;sample= annotations) against a set of OTU centroids:

vsearch \
    --usearch_global reads.fasta \
    --db otus.fasta \
    --id 0.97 \
    --otutabout otu_table.tsv

CITATION

Rognes T, Flouri T, Nichols B, Quince C, Mahé F. (2016) VSEARCH: a versatile open source tool for metagenomics. PeerJ 4:e2584 doi: 10.7717/peerj.2584

REPORTING BUGS

Submit suggestions and bug-reports at https://github.com/torognes/vsearch/issues, send a pull request on https://github.com/torognes/vsearch, or compose a friendly or curmudgeont e-mail to Torbjørn Rognes (torognes@ifi.uio.no).

AVAILABILITY

Source code and binaries are available at https://github.com/torognes/vsearch.

These manual pages are also published online at https://torognes.github.io/vsearch/.

COPYRIGHT

Contact: Torbjørn Rognes torognes@ifi.uio.no, Department of Informatics, University of Oslo, PO Box 1080 Blindern, NO-0316 Oslo, Norway

This software is dual-licensed and available under a choice of one of two licenses, either under the terms of the GNU General Public License version 3 or the BSD 2-Clause License.

GNU General Public License version 3

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

The BSD 2-Clause License

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

ACKNOWLEDGMENTS

We would like to thank the authors of the following projects for making their source code available:

vsearch includes code from Google’s CityHash project by Geoff Pike and Jyrki Alakuijala, providing some excellent hash functions available under a MIT license.
vsearch includes code derived from Tatusov and Lipman’s DUST program that is in the public domain.
vsearch includes public domain code written by Alexander Peslyak for the MD5 message digest algorithm.
vsearch includes public domain code written by Steve Reid and others for the SHA1 message digest algorithm.
vsearch binaries may include code from the zlib library, copyright Jean-Loup Gailly and Mark Adler.
vsearch binaries may include code from the bzip2 library, copyright Julian R. Seward.