Manual for vsearch 2.32.0, the current release. Changes that are not released yet are described in the development manual.

NAME

history — release history of vsearch

DESCRIPTION

This page is the release history of vsearch. It records the new features and the important modifications of each release, and nothing else; see vsearch(1) for the list of commands.

VERSION HISTORY

New features and important modifications of vsearch, most recent first (short-lived or minor bug releases may not be mentioned):

v2.32.0 released September 17th, 2026

Includes the following changes:

fix: --gapopen/--gapext * now forbid gaps at the match/merge decision, not only in the alignment score: a * opening penalty forbids that gap class outright, a * extension penalty forbids gaps longer than one, and both apply per sequence (Q/T) and location (L/I/R/E). For instance --gapopen '*LQ' forbids opening left end-gaps in the query, and a bare --gapopen '*' forbids every gap opening, including the terminal gaps used for substring matching. Previously * only lowered the score, so a pair whose best alignment still needed a forbidden gap was reported as a match (issue #602, and a 2016 forum report of internal gaps appearing despite --gapopen '*I').
fix: stop truncating the * infinite gap penalty on the SIMD path, so --gapopen/--gapext * forbid gaps on short sequences too, not just on long ones.
fix: stop truncating abundance to 32 bits when writing size annotations.
fix: stop truncating abundances above 2^32 in the --sortbysize and --sortbylength decks, where the most abundant sequences sorted to the bottom (;size=4294967297 truncated to 1, ranking below ;size=100) and “Median abundance:” reported a median of the truncated values.
fix: --fastq_mergepairs with --fastq_ascii 64 and --fastq_qmaxout between 63 and 93 wrote quality symbols outside printable ASCII (byte 127 at qmaxout 63, up to 149 at 93), producing an invalid FASTQ file, and a SIGSEGV in v2.31.0, which indexes a table with the resulting negative char. The sum rule was stated against --fastq_asciiout, which is the one fastq-writing command that does not accept it. --fastq_qminout was rejected for the mirror-image reason and is now accepted where the real offset represents it.
fix: the k-mer pre-filter required a target to share more k-mers than it can hold, so a database sequence with less than wordlength + minwordmatches - 1 = 19 nt of distinct k-mer content could not be selected by any query, an exact match included, and the same pair gave opposite answers depending on direction (issue #328).
fix: --fastx_uniques --strand both --fastqout merged a minus-strand member’s quality string positionally against the plus-strand seed, averaging seed position i with member position i instead of seqlen - 1 - i. Output changes versus released binaries for strand-both fastq dereplication.
fix: --fastx_uniques --fastqout truncated merged qualities through an irreversible probability round trip: quality 2 came back as 1, and qualities 3, 5, 6, 8, 10 and 17 moved with the abundances (two identical Q10 reads merged to Q10 at abundances 1+1, but to Q9 at 1+2). Present since v2.20.0.
fix: --selfid never excluded anything on any chimera path. The gate compared a query part, always shorter than a full-length identical reference, so the length pre-check short-circuited, and an identical reference in the database won every smoothing window and suppressed detection entirely.
fix: --cluster_unoise silently ignored a user-supplied --weak_id, overwriting it with 0.90 unconditionally; 0.90 is now a true default.
fix: count --qsegout, --tsegout and --fastapairs as output files in the “No output files specified” gates of --allpairs_global, --usearch_global, --search_exact and the --cluster_* commands, so each can be used as the sole output. Same defect and same fix as issue #82.
fix: enforce the documented ranges of option arguments. A negative --idprefix/--idsuffix crashed (out-of-bounds read under NDEBUG); --query_cov/--target_cov outside 0.0-1.0 silently made the filter unsatisfiable or inert; negative --maxdiffs, --maxsubs, --maxgaps, --mincols, --maxqsize, --mintsize and --topn silently rejected every hit or ignored the option entirely; --maxrejects -1 was indistinguishable from not passing the option; and an empty argument such as --id "" was read as zero, accepting every hit.
fix: accept a correctly-rounded underflowing double such as --fastq_maxee 1e-320, which died with “Illegal option argument” although the manual admits any positive real. Overflow stays rejected. Whether strtod even sets ERANGE on underflow is implementation-defined, so the old rejection also varied with the platform’s libc.
fix: --fastx_getseqs --label_word "" matched every record. An empty needle gave std::strstr no exit condition, so the search walked past the end of the header into the zero-filled buffer tail.
fix: --fastq_filter/--fastx_filter silently ignored --fastq_truncqual on fasta input instead of rejecting it; the comparison tested < LONG_MIN, which no value satisfies, where != LONG_MIN was meant. The fatal message has always listed it among the not-accepted options.
fix: a second one-past-the-end abundance read in --fastx_subsample (random_subsampling), in the priming read before the loop, reached on an empty database. The read inside the loop was guarded by PR #650; this is the one four lines above it.
fix: --makeudb_usearch wrote a UDB declaring zero sequences and exited 0, then rejected that file itself on every subsequent read with “Invalid UDB file”. Reachable through --minseqlength, not only through an empty input file.
fix: extend the UDB load validation of PR #647 to --udbinfo, which reads the header through its own path and so accepted both a UDB served from a pipe and an inflated stored sequence count, printing a garbage count and exiting 0. A stored word count larger than the database is now rejected by every reader too: it cannot describe a valid word list, and such slots were never visited while still being reported.
fix: --udb2fasta, --udbstats and --udbinfo no longer print “WARNING: Wordlength adjusted to N as indicated in UDB file”. None of the three accepts --wordlength, so the warning announced a change to a setting the user had no way of making; for --udbstats the report prints the word length two lines later anyway. Search commands, which do have a configured value for the file to override, keep it.
fix: uppercase the reference bases in SAM MD fields, as the SAMtags grammar requires. A target lowercased by DUST masking produced MD:Z:20a19 and MD:Z:20^a19, which disagree with what samtools calmd recomputes for the same alignment, and with the file itself after a BAM round trip.
fix: --sff_convert read 32 header bytes where the format specifies 31 (the struct carries a trailing padding byte), shifting every later section by one: the reported key sequence lost its first base, and spec-valid files were wrongly rejected whenever 31 + flows + key is a multiple of 8. The per-record offsets re-synchronized through the padding arithmetic, so the fastq output itself was always correct.
fix: --gzip_decompress no longer disables the transparent read that lets uncompressed standard input through. Feeding plain input to the option died with “incorrect header check” instead of reading it as-is, against zlib 1.3.2 as shipped by FreeBSD.
fix: --fastq_stats searched past the end of its per-position tables for the first complete EE filtering row. The bound is len_max, as in the pre-refactoring report loop.
fix: --fastq_eestats2 printed “0 reads, max len 0, avg -nan” to --log on empty input, where --output printed “0 reads”.
fix: restore one progress line per percent step when standard error is not a terminal. The behaviour added in v2.31.0 was silently reverted when the callers migrated to the progress class, which had been written from the older code.
fix: warnings now reach the --log file. Two of the fifteen warning sites (--sff_convert’s missing index padding, the UDB adjusted-wordlength notice) wrote to standard error and stopped there, and the --rereplicate and sequence-retrieval warnings obeyed --quiet on standard error while logging unconditionally. The manual says --quiet suppresses everything “except for warnings and error messages”.
fix: --chimeras_parents_max above 20 was an out-of-bounds heap write, and --iddef outside 0-4 selected an undefined identity definition, for a caller reaching the engines through the library: both were range-checked in the command-line parser only. Both moved to the shared validator, and the parents_found <= 20 assert became a runtime check so it holds under NDEBUG.
fix: reject - given as both the query and --db. Since --db gained standard-input support the two readers raced over the same stream and the command silently returned no hits with exit 0 (issue #506).
fix: --orient against a database of fewer than eight sequences gave a bitmap to every one of the 4^wordlength index slots, absent k-mers included, because integer division makes the threshold zero: +512 MB of peak memory and a full SIMD bitmap scan per absent k-mer (0.85 s to 0.25 s on a 4-sequence database).
fix: undefined behaviour at twelve <cctype> call sites passing a plain char, reachable wherever char is signed (x86-64, Windows) through sequence headers carrying UTF-8 or Latin-1.
fix: --fastq_mergepairs no longer reports two discard reasons that could never occur: “indel errors”, unassigned since it was added in 2017 because an ungapped aligner cannot detect an indel, and “undefined reason”.
fix: the FASTQ offset-warning threshold counted 101 records where the constant and its comment both say 100.
fix: --help now lists the commands --fastx_getseq, --fastx_getseqs and --fastx_getsubseq, under a new “Sequence retrieval” section, along with the options specific to them (--label, --labels, --label_word, --label_words, --label_field, --label_substr_match, --subseq_start, --subseq_end and --notmatchedfq). The three commands have manual pages but could not be discovered from the help text.
fix: --help now lists the output annotation options --lengthout, --xlength and --sample wherever it already lists their siblings --sizeout and --xsize, lists --allow_fewer under --fastx_subsample, and lists --fastq_qout_max, the last of the twenty-nine options --fastx_uniques accepts that appeared nowhere in the help text.
fix: out-of-bounds read in --fastx_getsubseq when --subseq_start exceeds a sequence’s length, in mixed-length files (also affects the --fastqout path) (PR #650).
fix: one-past-the-end abundance read in --fastx_subsample (random_subsampling), reached with --sizein when the last selected read is the last read of the last amplicon (PR #650).
fix: hit-list buffer overflow in clustering with large --maxaccepts/--maxrejects on small datasets (PR #650).
fix: harden the SFF reader (--sff_convert) against truncated or corrupt files: inverted clip regions under --sff_clip, flowgram truncation, and overflow guards that were previously compiled out of release builds (PR #650).
fix: bound-check the CIGAR walk in SAM output (--samout) to prevent out-of-bounds reads on malformed or mismatched alignments (PR #649).
fix: reject non-finite floating-point command-line arguments such as --sample_pct nan or inf, which previously slipped past range checks (PR #649).
fix: on Windows, load the zlib and bzip2 DLLs with LOAD_LIBRARY_SEARCH_DEFAULT_DIRS so a planted zlib1.dll/libbz2.dll in the working directory can no longer be loaded on the first .gz/.bz2 input (PR #649).
fix: validate UDB header fields on load (kmerindex entries, kmerindexsize, seqcount, header offsets); corrupt or crafted .udb files are now rejected with “Invalid UDB file” instead of causing out-of-bounds memory access; valid files are unaffected (PR #647).
fix: avoid int overflow when sizing and indexing the --fastq_eestats2 count table with a large --length_cutoffs range and many --ee_cutoffs values (PR #644).
fix: avoid intermittent crashes from terminating in a worker thread on malformed or mismatched input (PR #635, #636).
fix: data race on the query file position in --uchime_ref with multiple threads (PR #637).
fix: print FASTA sequences longer than INT_MAX (about 2 GB, reachable with a raised --maxseqlength) correctly instead of silently truncating them, and reject FASTA/FASTQ headers longer than the int length limit instead of overflowing the query buffer (PR #643).
fix: bound sequence length at the central FASTA/FASTQ reader and lower the --maxseqlength hard cap to INT_MAX - 2001 2147481646) to prevent int overflow in the search, cluster and chimera engines (PR #646; refines the 2^32-1 cap from PR #632).
fix: detect write, flush and close errors on output streams; a deferred write failure (full disk, exceeded quota, read-only remount) now aborts with a fatal error instead of silently producing a truncated output file that exits 0 (PR #651).
fix: duplicated annotation separators (;) in fastq and fasta headers (issue #271).
fix: shared-kmer counter overflow when aligning very long sequences (about 75 kbp) (PR #630).
fix: SIMD alignment counters (aligned/matches/mismatches/gaps) could overflow when the alignment path exceeds 65535 columns; over-long sequence pairs are now diverted to the linear-memory aligner, with 64-bit length arithmetic (PR #631).
fix: handle sequence abundances above 2^31 (;size=) in storage, filtering and output (including ;seqs= and centroid size) and in the library API; the abskew/size-ratio comparison is now exact above 2^53, and --maxseqlength rejects values above 2^32-1 (PR #632).
fix: commands --fastq_filter, --fastx_filter, --fastq_eestats and --fastq_mergepairs write qmin to the log file instead of standard error.
fix: off-by-one in MSA consensus ;length= output (PR #629).
fix: command --rereplicate writes its warning to the log file instead of standard error (PR #628).
add: new command --scramble: randomize the nucleotide order within each fasta or fastq entry, the within-entry counterpart of --shuffle. Useful for building null-model datasets: entry order, headers, lengths, and per-entry nucleotide composition are preserved; fastq quality strings are copied through unchanged. Reproducible with --randseed. The option --scramble_kmer k (1 to 9, default 1) additionally preserves the counts of all words of length k or less, by sampling a uniformly random Eulerian path of each entry’s de Bruijn graph (Altschul-Erickson/uShuffle-style).
add: ten new --userfields output fields (issue #548): diffs and mid, which report the quantities --maxdiffs and --mid are compared against; qseq and tseq, the full-length query and target sequences; qrowdots and trowdots, the aligned segments with a dot at every position identical to the other sequence; and qlor, qhir, tlor and thir, the alignment span in 0-based coordinates.
add: new option --allow_fewer for --fastx_subsample, which reads --sample_size as an upper bound: take at most N reads, and keep them all when fewer are available. Without it, a subset size larger than the input remains an error, as in usearch.
add: new option --fastq_solexa for --fastq_convert, which converts Solexa scores (-10 log10(p / (1 - p))) to Phred scores. The two scales share the ASCII offset 64 and nothing else, so a Solexa file was previously rebased and left otherwise untouched: fifteen of the 68 representable symbols came out wrong, and they were exactly the low ones a quality filter acts on.
add: a warning when the quality symbols read contradict --fastq_ascii. Raising --fastq_qmax means a phred+64 file read at the default offset 33 is no longer stopped by the bound, so vsearch now says so instead. Run --fastq_chars to confirm.
add: a warning when database sequences yield no k-mer for the index (issue #570). Such sequences are absent from the index and are never selected as candidates, so an all-lowercase reference database silently produced empty classifications. Emitted once per run as a summary line, not once per sequence.
add: a warning when a minimum threshold exceeds its matching maximum (--minuniquesize/--maxuniquesize, --minseqlength/--maxseqlength). Such a pair selects nothing, and the command used to write an empty output file and exit 0; the empty output stays legal. Also catches the implicit inversion, for instance --maxseqlength 10 alone against the default minimum of
add: a warning when --sample resolves to an empty sample name, which writes a bare ;sample= annotation and yields an OTU table whose first column has no name.
add: a warning when --gzip_decompress or --bzip2_decompress contradicts a seekable standard input, where autodetection runs and wins and the asserted format is ignored. On piped standard input the same mistake is already fatal.
add: shell auto-completion scripts for bash, zsh and fish (issue #417), installed by make install in the standard location for each shell. Completion offers the commands, then only the options the chosen command accepts, values for options taking a fixed set, and filenames filtered by extension. Configure with --disable-completion or --with-bashcompdir=DIR and friends.
add: an online manual, published per command, with the released pages at /vsearch/ and the development pages under /vsearch/dev/. --help now points at both.
add: a --enable-sanitize configure switch that builds the vsearch executable under the address and undefined-behaviour sanitizers, independently of --enable-debug.
add: new command --fastx_syncpairs.
add: preliminary support for FreeBSD/x86-64 (PR #633, #634).
add: new option --centroid_sizeout, which reports the centroid size in the FASTA header after clustering (issue #623).
remove: --search_exact no longer accepts --lcaout and --lca_cutoff. It never implemented them, in this tree or upstream: no call to the LCA writer exists on that code path, so the run completed, exited 0, and created no file. --usearch_global, the only command that implements LCA output, is unaffected.
remove: the monolithic vsearch.1 manual page, deprecated since the per-command pages and the online manual took over.
change: --fastq_qmax and --fastq_qmaxout now default to the highest quality score the offset can represent (93 with --fastq_ascii 33, 62 with offset 64) instead of 41. PacBio HiFi reaches Q93 and nanopore basecallers pass Q50, and the old bound rejected those files outright (issues #522, #194 and #609). Pass --fastq_qmax 41 for the previous behaviour.
change: --fasta2fastq and --fastq_mergepairs keep the previous --fastq_qmaxout default of 41, because they generate the score they clamp rather than passing one through: --fasta2fastq has no input quality, and --fastq_mergepairs caps a computed posterior (two agreeing Q40 bases would otherwise be reported as Q85 instead of Q41). Pass --fastq_qmaxout 93 for the unclamped values.
change: the default thread count, and an explicit --threads 0, now follow the cores this run may actually use, namely the CPU affinity mask and the cgroup CPU quota, instead of the machine’s core count; memory detection is cgroup-aware for the same reason. Inside a Slurm, Docker, podman, Kubernetes or Apptainer confinement vsearch read the host’s figures: a session limited to 50 MB and one core was told it had 500+ GB and 128 cores, and then really did start 128 workers on that one core (issue #584). A virtual machine was already correct and is unaffected.
change: --fastq_stats now requires --log, as the manual documents; the report has no other destination.
change: --fastq_mergepairs now honours --quiet for its final report on standard error, diverging from the issue #527 resolution, and --log no longer takes that report away from standard error: every other command writes both copies.
change: --fastq_mergepairs applies --fastq_maxlen after quality truncation, as the shared option fragment (“Applied after trimming”) and the discard-reason string (“reads too long (after truncation)”) both already stated. A pair truncated to within --fastq_maxlen is no longer discarded as too long.
change: --fastx_uniques now applies --fastq_qmin and --fastq_qmax to its input. Both were listed among its valid options and neither was enforced, although the command decodes every quality symbol, converts it to an error probability, takes an abundance-weighted mean across the cluster and re-encodes the result.
change: --derep_prefix now requires an output option, like --derep_fulllength and --derep_id. A run with neither --output nor --uc used to read the input, write nothing, and exit 0.
change: --fastq_filter and --fastx_filter now reject a *_rev output given without --reverse, for which the file was never opened: such a run created no file, wrote nothing, and exited 0.
change: --udb2fasta --sizeout now needs --sizein to reproduce the abundances stored in the UDB, following the convention of every other command. Previously --sizeout alone overwrote every abundance with size=1, so the option that asks for the annotation was the one destroying it, and --sizein had no effect on this path.
change: --makeudb_usearch masks on several threads instead of being forced to one and warning. Only the masking phase is parallel; reading, indexing and writing the UDB remain serial.
change: four commands accept options they already needed. --cut accepts --threads (it warns and forces one thread, like every other single-threaded command); --fastx_mask, --fastx_subsample and --orient accept --minseqlength/--maxseqlength, whose input was silently capped at the 50000-nt default with no way to raise it; --derep_fulllength accepts --label_suffix, like its four sibling dereplication commands; and --fastx_uniques accepts --tabbedout with fasta input, since none of the six tabbedout columns depends on quality data.
change: the --sintax --randseed multithreading warning is removed. The per-query RNG is seeded from a per-query substream, so the classification is reproducible whatever the thread count, exactly as the manual page states.
change: reject --match/--mismatch outside -32767..32767 and finite --gapopen/--gapext outside 0..6553, instead of silently wrapping them to the 16-bit aligner.
improve: --fastq_filter and --fastx_filter are about 2x faster (552 ms to 282 ms on 300000 records, byte-identical). Most of the gain is in the shared FASTQ reader, so every FASTQ-reading command benefits: --fastq_stats -27%, --fastq_eestats2 -15%, --fastq_join -13%, --fastq_convert -13%, --fastx_revcomp -12%, --fastx_subsample -11%, --fastq_chars -11%, --fastx_uniques -8%.
improve: --fastq_mergepairs is 3.0x faster on one thread and 1.9x on eight, and 3.5x on a non-overlapping pair set, chiefly by indexing its 5-mers directly (1024 possible values) instead of hashing them.
improve: --search_exact is 1.74x faster at --threads 24 and uses 61 MB instead of 3752 MB, by growing the per-thread hit buffer on demand rather than sizing it to the whole database, per strand, per thread. That allocation, not the search, was what made the command slower with more threads.
improve: --sintax is 15% to 38% faster depending on how query-heavy the run is: byte-wide hit counters (a sintax count cannot exceed 32, so a 16-bit counter is twice the width it needs) and a block skip over the counters that cannot beat the running best, which is 99.94% of them.
improve: --udb2fasta is 7% to 82% faster and uses 21 MB instead of 1.0 GB on a word length 13 database, by not loading a k-mer index it never reads. --udbstats no longer sorts the whole k-mer table to print eleven rows and two order statistics (up to -91%), and no longer builds that report when --log is absent and it cannot be printed.
improve: --makeudb_usearch is 35% faster (4.93 s to 3.18 s on the PR2 reference): parallel masking, and a word-at-a-time scan that expands a bitmapped k-mer 8.6x faster than bit by bit.
improve: --uchime_denovo and the --uchime*_ref commands are 13% faster: clearing only the live prefix of the match and insert arrays (about 9 rows of the 400 allocated) and wiping each match column once instead of once per overlapping window.
improve: --fastx_syncpairs executes 16% fewer instructions and allocates far less, storing the reverse reads in a chunked arena instead of three separate allocations per record.
improve: --fastx_getseqs --labels hashes the labels for exact whole-header matching instead of comparing every record with every label: 1.95 billion string comparisons for 10000 labels against 200000 records, 90% of the command’s runtime, and quadratic growth on both axes.
improve: DUST masking skips the start positions that provably cannot score, which is 99.4% of windows on real data.
improve: --fastq_eestats can now process long reads in a memory-efficient way. It is also faster (14-27% in our tests) when processing short reads.
improve: output-open failure messages are harmonized and now include the offending filename (for instance --alnout, --samout, --uc, --clusters) (PR #652).
improve: building no longer requires ./autogen.sh or the autotools; the generated build files are committed and authoritative (maintainer mode disabled), so an ordinary build needs only a C++ compiler and make (PR #648).
improve: recommend -O3 again now that the SIMD-aligner strict-aliasing miscompile (issue #589) is fixed at the root and the -fno-tree-partial-pre workaround is removed (PR #649).
improve: a documentation audit of the whole manual, correcting the per-command pages against measured behaviour: the actual --minseqlength default (32) for four commands, --derep_prefix accepting only --strand plus, --borderline as a sole output option for the uchime commands, --sizeout propagating rather than resetting sizes, how alignments against Ns are scored and counted, the OTU-table identifier uniqueness requirement, the cigar(5) example and the SAM run-length exception, --fastq_solexa, --derep_fulllength marked deprecated in favour of --fastx_uniques, and the accepted ranges for --match/--mismatch and the gap penalties.
improve: fix double-hyphen rendering of option names in the manual pages and the online HTML documentation, including the vsearch-userfields(7) NAME line.
improve: cross-platform pseudo-random number generator reproducibility (PR #640).
improve: documentation (issues #239, #282).
improve: code testing (new automatic tests in vsearch-tests; PR #638, #641).

v2.31.0 released April 29th, 2026

Includes the following changes:

fix: --fastx_getseqs --label_words out-of-bounds memory issue
fix: --fastx_getsubseq now writes non-matching sequences in full (no trimming)
fix: sintax’s tie-breaking mechanism
fix: limit the scope of decompress options (--gzip_decompress and --bzip2_decompress) to stdin
fix: option --sample now properly truncates at first ‘;’ or blank
fix: option --wordlength now checks for values outside the range 3-15
fix: option --fasta_width now checks for negative values
fix: option --maxseqlength now checks for null and negative values
change: restore progress indicator when stderr is not a terminal. One line per progress step. This changes behavior present since v1.10.0.
change: --fastq_mergepairs now honors --sizeout and appends ;size=1 to merged and not-merged output headers
change: --fastq_filter and --fastx_filter: now reject out-of-domain values for --fastq_maxee, --fastq_maxee_rate, --fastq_truncee, --fastq_maxlen, --fastq_maxns, --fastq_minlen, --fastq_trunclen, --fastq_trunclen_keep, --fastq_truncqual, --fastq_stripleft, and --fastq_stripright
change: option --eetabbedout now prints EE values with more digits (more precise values)
change: --fastx_getseqs --label_words now silently skips empty lines
add: online documentation for all vsearch commands (https://torognes.github.io/vsearch/)
add: command --orient now supports option --hardmask
add: compilation flag -fno-exceptions
add: static library API for vsearch (contributed by @wasade)
improve: code refactoring
improve: code coverage of our test-suite (8,930 active tests)

v2.30.6 released March 27th, 2026

Includes the following changes:

fix: out-of-bound look-ups when printing empty last common ancestor (LCA) results with the --lcaout option. Thanks to user AntoninLCH for reporting this issue,
improve: code refactoring,
improve: code coverage of our test-suite.

v2.30.5 released March 10th, 2026

Includes the following changes:

fix: out-of-bound look-ups when printing alignment rows (commit 2f3d1ef). Thanks to user @gbbio for reporting this issue (#618). This is a regression, introduced in v2.30.1,
fix: options --xlength and --xee were not removing length and ee attributes as expected (commit 69fccdbc). This is a regression, introduced in v2.30.1,
fix: range checking for the option --max_unmasked_pct (commit e1a5611505),
add: individual manpages for the commands --shuffle and --cut,
improve: eliminated 50 clang-tidy warnings,
improve: code coverage of our test-suite.

v2.30.4 released January 19th, 2026

Fixes issue #617. Due to a bug, incorrect scores were reported in the FASTA file headers of some non-chimeric sequences during chimera detection. The non-chimeric status was correct.

v2.30.3 released January 12th, 2026

Fixes memory allocation bugs in the chimera detection code that caused a segmentation fault in rare cases (issue #615).

v2.30.2 released December 12th, 2025

Fixes two minor issues. Allow a UDB file to be written to a pipe or stdout (issue #599). Correct computation of median cluster size after dereplication (issue #611).

v2.30.1 released October 3rd, 2025

Incorporates many code improvements, more extensive testing, better documentation and some minor bug fixes:

fix: use-after-free introduced in commit de6c1d8 (Jun 13, 2024),
fix: (harmless) out-of-bounds memory issue in --derep_prefix (commit 8a0a508b),
fix: (harmless) memory leak in --fastx_getseqs --label_field (commit a9c42713),
fix: (harmless) memory leak when using option --userfields (--allpairs_global commit 03b95bcf; --cluster_* commit 2fde5472; --search_exact commit 45cd56d6; --usearch_global commit d83bfee9),
fix: (harmless) valgrind error, use of uninitialized values (commit 8bab2444), also eliminates a pesky compilation warning,
change: passing a negative value to --fastq_truncee_rate is now an error (commit a120f371),
change: passing a negative value to --fastq_minqual is now an error (commit ff5b0c99),
change: passing a non-ASCII symbol to --join_padgap or --join_padgapq is now an error (commit a708f5b3),
change: when using --gapopen "*" to forbid gap opening, the penalty is now set to INT_MAX (rather than 1,000). This might change alignment results for users who relied on the old behavior (thanks to Denis Filloux, issue #602, commit 96e9cf9e),
change: when using command --chimeras_denovo, --tabbedout or --alnout can be the only output files specified (commit d51f0a4),
change: when using command --chimeras_denovo, option --lengthout is now accepted (commit 3f55cc6b),
change: when using command --chimeras_denovo, option --xlength is now accepted (commit 0fd346cd),
add: compilation option GLIBCXX_DEBUG when compiling for debugging (commit 5cf4a6c1),
add: official support for clang 20,
add: initial support for clang 21, initial support for GCC 15,
add: experimental support for clang 22,
improve: more accurate line number when reporting illegal characters in fastq headers (commit 539084e9),
improve: more accurate line number when reporting non-ASCII characters in fastq headers (commit 98a851ed),
improve: remove checks for unneeded libraries during compilation (commit 249bb5d5),
improve: code quality (8,536 clang-tidy warnings eliminated),
improve: documentation and help messages (issue #604),
improve: complete refactoring and modernization of the command --fastq_stats,
improve: command --fastq_stats is now up to twice faster (tested on x86-64),
improve: extensive test-suites for --fastq_stats and --sff_convert,
improve: code coverage of our test-suite.

v2.30.0 released February 27th, 2025

Add options --n_mismatch, --fastq_minqual, and --fastq_truncee_rate. The --n_mismatch option will count N’s as mismatches in alignments, which may be useful to get sensible alignments for sequences with lots of N’s. By default N’s are counted as matches. Both the scoring and the counting of matches are affected. The new --fastq_minqual option for the --fastq_filter and --fastx_filter commands will discard sequences with any bases with a quality score below the given value. The new --fastq_truncee_rate option for the same commands will truncate sequences at the first position where the number of expected errors per base is above the given value.

v2.29.4 released February 14th, 2025

Adjust the window size used for chimera detection down from 64 to 32. The window size was by accident increased from 32 to 64 in version 2.23.0, leading to somewhat fewer chimeras being predicted. In addition, a compiler pragma has been included in align_simd.cc to further protect the compiler from generating wrong code.

v2.29.3 released February 3rd, 2025

Released to mitigate a bug that occurs when compiling the align_simd.cc file on x86_64 systems with the GNU C++ compiler version 9 or later with the -O3 optimization option. It results in incorrect code that may cause bad alignments in some circumstances. We are investigating this issue further, but for now we recommend compiling with the -O2 flag. The README.md file and the Dockerfiles have been updated to reflect this. The binaries released with this version will include this fix.

v2.29.2 released December 20th, 2024

Fix a segmentation fault during clustering when the set of clusters is empty. Initial documentation in markdown format available on GitHub Pages.

v2.29.1 released October 24th, 2024

Fix a segmentation fault that could occur during alignment in version 2.29.0, for example with --uchime_ref. Some improvements to code and documentation.

v2.29.0 released September 26th, 2024

Fixes seven bugs (see changelog below), adds initial support for RISC-V architectures, and improves code quality and code testing (1,210 new tests):

add: experimental support for RISCV64 and other 64-bit little-endian architectures, thanks to Michael R. Crusoe and his fellow Debian developers (issue #566),
add: official support for clang-19 and gcc 14,
add: beta support for clang-20,
remove: unused --output option for command --fastq_stats (issue #572),
fix: bug in --sintax when selecting the best lineage (only low confidence values below 0.5 were affected) (issue #573),
fix: out-of-bounds error in --fastq_stats when processing empty reads (issue #571),
fix: bug in --cut, patterns with multiple cutting sites were not detected (commit 4c4f9fa),
fix: memory error (segmentation fault) when using --derep_id and --strand (issue #565),
fix: --fastq_join now obeys --quiet and --log options (commit 87f968b),
fix: --fastq_join quality padding is now also set to Q40 when quality offset is 64 (commit be0bf9b),
fix: (partial) --fastq_join’s handling of abundance annotations (commit f2bbcb4),
improve: additional safeguards to validate input values and to make sure that they are within acceptable limits. Changes concern options --abskew (commit a530dd8) and --fastq_maxdiffs (commit 4b254db),
improve: code quality (1.3k+ commits, 6k+ clang-tidy warnings eliminated),
improve: documentation and help messages (issue #568),
improve: complete refactoring and modernization of a subset of commands (--sortbylength, --sortbysize, --shuffle, --rereplicate, --cut, --fastq_join, --fasta2fastq, --fastq_chars),
improve: code coverage of our test-suite for the above-mentioned commands (1,210 new tests, 4,753 in total).

v2.28.1 released April 26th, 2024

Fix a segmentation fault that could occur with the blast6out and output_no_hits options.

v2.28.0 released April 26th, 2024

The sintax command has been improved in several ways in this version of vsearch. Please note that several details of this algorithm are not clearly described in the preprint, and the implementation in vsearch differs from that in usearch. The former vsearch version did not always choose the most common taxonomic entity over the 100 bootstraps among the database sequences with the highest amount of word similarity to the query. Instead, if several sequences had an equal similarity with the query, the sequence encountered in the earliest bootstrap was chosen. The confidence level was calculated based on this sequence compared to the selected sequences from the other 99 bootstraps. This could lead to a suboptimal choice with a low confidence. In the new version, the most common of the sequences with the highest amount of word similarity across the 100 bootstraps will be selected, and ties will be broken randomly. Another problem with the old implementation was that if several sequences had the same amount of word similarity, the shortest one in the reference database would be chosen, and if they were equally long, the earliest in the database file would be chosen. A new option called sintax_random has now been introduced. This option will randomly select one of the sequences with the highest number of shared words with the query, without considering their length or position. This avoids a bias towards shorter reference sequences. This option is strongly recommended and will probably soon be the default. Furthermore, a ninth taxonomic rank, strain (letter t), is now recognized. The speed of the sintax command has also been significantly improved at least in some cases. Run vsearch with the randseed option and 1 thread to ensure reproducibility of the random choices in the algorithm.

v2.27.1 released April 6th, 2024

Fixes the weak_id option and makes searches report weak hits in some cases. Updates the names of the compression libraries to libz.so.1 and libbz2.so.1 on Linux to make them work on common Linux distributions without installing additional packages. README.md has been updated with information about compression libraries on Windows.

v2.27.0 released January 19th, 2024

The usearch_global and search_exact commands now support FASTQ files as well as FASTA files as input. This version of vsearch includes clarifications and updates to the manual. Some code has been refactored. Generic Dockerfiles for major Linux distributions have been included. Some warnings from compilers and other tools have been eliminated. The release for Windows will also include DLLs for the two compression libraries.

v2.26.1 released November 25th, 2023

No real changes, but the previous version was released without proper updates to the source code.

v2.26.0 released November 24th, 2023

Enable the maxseqlength and minseqlength options for the chimera detection commands. When the usearch_global or search_exact commands are used, OTU tables will include samples and OTUs with no matches.

v2.25.0 released November 10th, 2023

Allow a given percentage of mismatches between chimeras and parents for the experimental chimeras_denovo command.

v2.24.0 released October 26th, 2023

Update documentation. Improve code. Allow up to 20 parents for the undocumented and experimental chimeras_denovo command. Fix compilation warnings for sha1.c. Compile for release (not debug) by default.

v2.23.0 released July 7th, 2023

Update documentation. Add citation file. Modernize and improve code. Fix several minor bugs. Fix compilation with GCC 13. Print stats after fastq_mergepairs to log file instead of stderr. Handle sizein option correctly with dbmatched option for usearch_global. Allow maxseqlength option for makeudb_usearch. Fix memory allocation problem with chimera detection. Add lengthout and xlength options. Increase precision for eeout option. Add warning about sintax algorithm, random seed and multiple threads. Refactor chimera detection code. Add undocumented experimental long_chimeras_denovo command. Fix segfault with clustering. Add more references.

v2.22.1 released September 19th, 2022

Fix compiler warning.

v2.22.0 released September 19th, 2022

Add the derep_smallmem command for dereplication using little memory.

v2.21.2 released September 12th, 2022

Fix problems with the lcaout option when using maxaccepts above 1 and either lca_cutoff below 1 or with top_hits_only enabled. Update documentation. Update code to avoid compiler warnings.

v2.21.1 released January 18th, 2022

Fix a problem with dereplication of empty input files. Update Altivec code on ppc64le for improved compiler compatibility (vector->__vector).

v2.21.0 released January 12th, 2022

Adds the sample, qsegout and tsegout options. Enables the use of UDB databases with uchime_ref.

v2.20.1 released January 11th, 2022

Fixes a bug in fastq_mergepair that caused an occasional hang at the end when using multiple threads.

v2.20.0 released January 10th, 2022

Added the fastx_uniques command and the fastq_qout_max option for dereplication of FASTQ files. Some code cleaning.

v2.19.0 released December 21st, 2021

Added the lcaout and lca_cutoff options to enable the output of last common ancestor (LCA) information about hits when searching. The randseed option was added as a valid option to the sintax command. Code improvements.

v2.18.0 released August 27th, 2021

Added the fasta2fastq command. Fixed search bug on ppc64le. Fixed bug with removal of size and ee info in uc files. Fixed compilation errors in some cases. Made some general code improvements. Updated manual.

v2.17.1 released June 14th, 2021

Modernized code. Minor changes to help info.

v2.17.0 released March 29th, 2021

The fastq_mergepairs command has been changed. It now allows merging of sequences with overlaps as short as 5 bp if the --fastq_minovlen option has been adjusted down from the default 10. In addition, much fewer pairs of reads should now be rejected with the reason ‘multiple potential alignments’ as the algorithm for detecting those have been changed.

v2.16.0 released March 22nd, 2021

Adds the orient command. Handles empty input files properly. Documentation has been updated.

v2.15.2 released January 26th, 2021

No real functional changes, but some code and compilation changes. Compiles successfully on macOS running on Apple Silicon (ARMv8). Binaries available. Code updated for C++11. Minor adaptations for Windows compatibility, including the use of the C++ standard library for regular expressions. Minor changes for compatibility with Power8. Switch to C++ header files.

v2.15.1 released October 28th, 2020

Fix for dereplication when including reverse complement sequences and headers. Make some extra checks when loading compression libraries and add more diagnostic output about them to the output of the version command. Report an error when fastx_filter is used with FASTA input and options that require FASTQ input. Update manual.

v2.15.0 released June 19th, 2020

Update manual and documentation. Turn on notrunclabels option for sintax command by default. Change maxhits 0 to mean unlimited hits, like the default. Allow non-ascii characters in headers, with a warning. Sort centroids and uc too when clusterout_sort specified. Add cluster id to centroids output when clusterout_id specified. Improve error messages when parsing FASTQ files. Add missing fastq_qminout option and fix label_suffix option for fastq_mergepairs. Add derep_id command that dereplicates based on both label and sequence. Remove compilation warnings.

v2.14.2 released January 28th, 2020

Fixed some issues with the cut, fastx_revcomp, fastq_convert, fastq_mergepairs, and makeudb_usearch commands. Updated manual.

v2.14.1 released September 18th, 2019

Fixed bug with sequences written to file specified with fastaout_rev for commands fastx_filter and fastq_filter.

v2.14.0 released September 11th, 2019

Added relabel_self option. Made fasta_width, sizein, sizeout and relabelling options valid for certain commands.

v2.13.7 released September 2nd, 2019

Fixed bug in consensus sequence introduced in version 2.13.0.

v2.13.6 released July 2nd, 2019

Added info about cut command to output of help command.

v2.13.5 released July 2nd, 2019

Added cut command to fragment sequences at restriction sites. Silenced output from the fastq_stats command if quiet option was given. Updated manual.

v2.13.4 released May 10th, 2019

Added information about support for gzip- and bzip2-compressed input files to the output of the version command. Adapted source code for compilation on FreeBSD and NetBSD systems.

v2.13.3 released April 30th, 2019

Fixed bug in FASTQ parsing introduced in version 2.13.2.

v2.13.2 released April 30th, 2019

Fixed bug related to improper handling of newlines on Windows. Allowed option strand plus to uchime_ref for compatibility.

v2.13.1 released April 26th, 2019

Minor changes to the allowed options for each command. All commands now allow the log, quiet and threads options. If more than 1 thread is specified for commands that are not multi-threaded, a warning will be issued. Minor changes to the manual.

v2.13.0 released April 11th, 2019

Added the --fastx_getseq, --fastx_getseqs and --fastx_getsubseq commands to extract sequences from a FASTA or FASTQ file based on their labels. Improved handling of ambiguous nucleotide symbols. Corrected behaviour of --uchime_ref command with options --self and --selfid. Strict detection of illegal options for each command.

v2.12.0 released March 19th, 2019

Take sequence abundance into account when computing consensus sequences or profiles after clustering. Warn when rereplicating sequences without abundance info. Guess offset 33 in more cases with fastq_chars. Stricter checking of option arguments and option combinations.

v2.11.1 released February 28th, 2019

Minor change to the handling of the weak_id and id options when using cluster_unoise.

v2.11.0 released February 13th, 2019

Added ability to trim and filter paired-end reads using the reverse option with the fastx_filter and fastq_filter commands. Added --xee option to remove ee attributes from FASTA headers. Minor invisible improvement to the progress indicator.

v2.10.4 released January 4th, 2019

Fixed serious bug in x86_64 SIMD alignment code introduced in version 2.10.3. Added link to BioConda in README. Fixed bug in fastq_stats with sequence length 1. Fixed use of equals symbol in UC files for identical sequences with cluster_fast.

v2.10.3 released December 19th, 2018

Ported to Linux on ARMv8 (aarch64). Fixed compilation warning with gcc version 8.1.0 and 8.2.0.

v2.10.2 released December 10th, 2018

Fixed bug in sintax with reversed order of domain and kingdom.

v2.10.1 released December 7th, 2018

Improved sff_convert command. It will now read several variants of the SFF format. It is also able to read from a pipe. Warnings are given if there are minor problems. Error messages have been improved. Minor speed and memory usage improvements.

v2.10.0 released December 6th, 2018

Added the sff_convert command to convert SFF files to FASTQ. Added some additional option argument checks. Fixed segmentation fault bug after some fatal errors when a log file was specified.

v2.9.1 released October 29th, 2018

Changed compiler options that select the target cpu and tuning to allow the software to run on any 64-bit x86 system, while tuning for more modern variants. Avoid illegal instruction error on some architectures. Update documentation of rereplicate command.

v2.9.0 released October 10th, 2018

Added the fastq_join command.

v2.8.6 released October 9th, 2018

Fixed a bug introduced in version 2.8.2 that caused derep_fulllength to include the full FASTA header in its output instead of stopping at the first space (unless the notrunclabels option is in effect).

v2.8.5 released September 26th, 2018

Fixed a bug in fastq_eestats2 that caused the values for large lengths to be much too high when the input sequences had varying lengths.

v2.8.4 released September 3rd, 2018

Further reduce memory requirements for dereplication when not using the uc option. Fix output during subsampling when quiet or log options are in effect.

v2.8.3 released August 31st, 2018

Fix for segmentation fault for --derep_fulllength with --uc.

v2.8.2 released August 21st, 2018

Fix for wrong placement of semicolons in header lines in some cases when using the sizeout or xsize options. Reduced memory requirements for full-length dereplication in cases with many duplicate sequences. Improved wording of fastq_mergepairs report. Updated manual regarding use of sizein and sizeout with dereplication. Changed a compiler option.

v2.8.1 released June 22nd, 2018

Fixes for compilation warnings with GCC 8.

v2.8.0 released April 24th, 2018

Added the fastq_maxdiffpct option to the fastq_mergepairs command.

v2.7.2 released April 20th, 2018

Added the sintax command for taxonomic classification. Fixed a bug with incorrect FASTA headers of consensus sequences after clustering.

v2.7.1 released February 16th, 2018

Fix several bugs on Windows related to large files, use of - as a file name to mean stdin or stdout, alignment errors, missed kmers and corrupted UDB files. Added documentation of UDB-related commands.

v2.7.0 released February 13th, 2018

Added commands cluster_unoise, uchime2_denovo and uchime3_denovo contributed by Davide Albanese based on Robert Edgar’s papers. Refactored fasta and fastq print functions as well as code for extraction of abundance and other attributes from the headers.

v2.6.2 released December 18th, 2017

Fixed option xsize that was partially inactive for commands uchime_denovo, uchime_ref, and fastx_filter.

v2.6.1 released December 8th, 2017

Improved parallelisation of paired end reads merging.

v2.6.0 released November 10th, 2017

Rewritten paired-end reads merger with improved accuracy. Decreased default value for fastq_minovlen option from 16 to 10. The default value for the fastq_maxdiffs option is increased from 5 to 10. There are now other more important restrictions that will avoid merging reads that cannot be reliably aligned.

v2.5.2 released October 30th, 2017

Fixed bug where - as an argument to the fastq_eestats2 option was treated literally instead of equivalent to stdin.

v2.5.1 released October 25th, 2017

Fixed bug with bad default value of 1 instead of 32 for minseqlength when using the makeudb_usearch command.

v2.5.0 released October 5th, 2017

Support for UDB database files. New commands: fastq_stripright, fastq_eestats2, makeudb_usearch, udb2fasta, udbinfo, and udbstats. New general option: no_progress. New options minsize and maxsize to fastx_filter. Minor bug fixes, error message improvements and documentation updates.

v2.4.4 released August 28th, 2017

Fixed a few minor bugs, improved error messages and updated documentation.

v2.4.3 released April 6th, 2017

Fixed bug with progress bar for shuffling. Fixed missing N-lines in UC files with usearch_global, search_exact and allpairs_global when the output_no_hits option was not specified.

v2.4.2 released March 10th, 2017

Default value for fastq_minovlen increased to 16 in accordance with help text and for compatibility with usearch. Minor changes for improved accuracy of paired-end read merging.

v2.4.1 released March 1st, 2017

Fixed an overflow bug in fastq_stats and fastq_eestats affecting analysis of very large FASTQ files. Fixed maximum memory usage reporting on Windows.

v2.4.0 released February 8th, 2017

Added support for Linux on Power8 systems (ppc64le) and Windows on x86_64. Improved detection of pipes when reading FASTA and FASTQ files. Corrected option for specifying output from fastq_eestats command in help text.

v2.3.4 released December 9th, 2016

Fixed bug in output of sequence profiles and updated documentation.

v2.3.3 released December 5th, 2016

A minor speed improvement.

v2.3.2 released November 18th, 2016

Fixed bug where vsearch reported the ordinal number of the target sequence instead of the cluster number in column 2 on H-lines in the uc output file after clustering. For search and alignment commands both usearch and vsearch reports the target sequence number here.

v2.3.1 released November 16th, 2016

Fixed bug where --minwordmatches 0 was interpreted as the default minimum word matches for the given word length instead of zero. When used in combination with --maxaccepts 0 and --maxrejects 0 it will allow complete bypass of kmer-based heuristics.

v2.3.0 released October 10th, 2016

Allowed zero-length sequences in FASTA and FASTQ files. Added --fastq_trunclen_keep option. Fixed bug with output of OTU tables to pipes.

v2.2.0 released October 7th, 2016

Added OTU table generation options --biomout, --mothur_shared_out and --otutabout to the clustering and searching commands.

v2.1.2 released September 28th, 2016

Fixed incorrect abundance output from fastx_filter and fastq_filter when relabelling.

v2.1.1 released September 23rd, 2016

Fixed bugs in output to UC-files. Improved help text and manual.

v2.1.0 released September 16th, 2016

New command: --fastx_filter. New options: --fastq_maxlen, --fastq_truncee. Allow --minwordmatches down to 3.

v2.0.5 released September 9th, 2016

Add options --fastaout_discarded and --fastqout_discarded to output discarded sequences from subsampling to separate files. Updated manual.

v2.0.4 released September 1st, 2016

Improved error message for bad FASTQ quality values. Improved manual.

v2.0.3 released August 2nd, 2016

Fixed bad compiler options resulting in Illegal instruction errors when running precompiled binaries.

v2.0.2 released July 5th, 2016

Avoid warnings when compiling with GCC 6.

v2.0.1 released June 30th, 2016

Avoid segmentation fault when masking very long sequences.

v2.0.0 released June 24th, 2016

This major new version supports reading from pipes. Two new options are added: --gzip_decompress and --bzip2_decompress. One of these options must be specified if reading compressed input from a pipe, but are not required when reading from ordinary files. The vsearch header that was previously written to stdout is now written to stderr. This enables piping of results for further processing. The file name - now represents standard input (/dev/stdin) or standard output (/dev/stdout) when reading or writing files, respectively. Code for reading FASTA and FASTQ files has been refactored.

v1.11.2 released June 21st, 2016

Two bugs were fixed. The first issue was related to the --query_cov option that used a different coverage definition than the qcov userfield. The coverage is now defined as the fraction of the whole query sequence length that is aligned with matching or mismatching residues in the target. All gaps are ignored. The other issue was related to the consensus sequences produced during clustering when only N’s were present in some positions. Previously these would be converted to A’s in the consensus. The behaviour is changed so that N’s are produced in the consensus, and it should now be more compatible with usearch.

v1.11.1 released April 13th, 2016

Added strand information in UC file for --derep_fulllength and --derep_prefix. Added expected errors (ee) to header of FASTA files specified with --fastaout and --fastaout_discarded when --eeout or --fastq_eeout option is in effect for fastq_filter and fastq_mergepairs. The options --eeout and --fastq_eeout are now equivalent.

v1.10.2 released March 18th, 2016

Fixed a bug causing a segmentation fault when running --usearch_global with an empty query sequence. Also fixed a bug causing imperfect alignments to be reported with an alignment string of = in uc output files. Fixed typos in man file. Fixed fasta/fastq processing code regarding presence or absence of compression library header files.

v1.10.1 released February 23rd, 2016

Fixed a bug affecting the --fastq_mergepairs command causing FASTQ headers to be truncated at first space (despite the bug fix release 1.9.0 of November 12th, 2015). Full headers are now included in the output (no matter if --notrunclabels is in effect or not).

v1.10.0 released February 11th, 2016

Parallelized and improved merging of paired-end reads and adjusted some defaults. Removed progress indicator when stderr is not a terminal. Added --fasta_score option to report chimera scores in FASTA files. Added --rereplicate and --fastq_eestats commands. Fixed typos. Added relabelling to files produced with --consout and --profile options.

v1.9.10 released January 25th, 2016

Fixed bug related to masking and lower case database sequences.

v1.9.9 released January 22nd, 2016

Adjusted default minimum number of word matches during searches for improved performance.

v1.9.8 released January 22nd, 2016

Fixed bug causing segfault when chimera detection is performed on extremely short sequences.

v1.9.7 released January 12th, 2016

Masking behaviour is changed somewhat to keep the letter case of the input sequences unchanged when no masking is performed. Masking is now performed also during chimera detection. Documentation updated.

v1.9.6 released January 8th, 2016

Fixed bug in aligned sequences produced with --fastapairs and --userout (qrow, trow) options.

v1.9.5 released December 3rd, 2015

Fixed bug resulting in inferior chimera detection performance.

v1.9.4 released December 3rd, 2015

Fixed incrementation of counter when relabeling dereplicated sequences.

v1.9.3 released November 19th, 2015

Workaround for missing x86intrin.h with old compilers.

v1.9.2 released November 17th, 2015

Fixed a bug in the computation of some values with --fastq_stats.

v1.9.1 released November 13th, 2015

Fixed memory leak and a bug in score computation in --fastq_mergepairs, and improved speed.

v1.9.0 released November 12th, 2015

Added the --fastq_mergepairs command and associated options. This command has not been tested well yet. Included additional files to avoid dependency of autoconf for compilation. Fixed an error where identifiers in fasta headers were not truncated at tabs, just spaces. Fixed a bug in detection of the file format (FASTA/FASTQ) of a gzip compressed input file.

v1.8.1 released November 2nd, 2015

Fixes for compatibility with QIIME and older OS X versions.

v1.8.0 released October 19th, 2015

Added --search_exact, --fastx_mask and --fastq_convert commands. Changed most commands to read FASTQ input files as well as FASTA files. Modified --fastx_revcomp and --fastx_subsample to write FASTQ files.

v1.7.0 released October 14th, 2015

Add --relabel_keep option.

v1.6.1 released October 14th, 2015

Fix bugs and update manual and help text regarding relabelling. Add all relabelling options to the subsampling command. Add the --xsize option to chimera detection, dereplication and fastq filtering commands. Refactoring of code.

v1.6.0 released October 9th, 2015

Adds the relabeling options (--relabel, --relabel_md5 and --relabel_sha1) to the shuffle command. Also adds the --xsize option to the clustering, dereplication, shuffling and sorting commands.

v1.5.0 released October 7th, 2015

Introduces the new option --minwordmatches that allows the user to specify the minimum number of matching unique words before a sequence is considered further. New default values for different word lengths are also set. The minimum word length is increased to 7.

v1.4.7 released October 7th, 2015

Fixes a problem with searching for or clustering sequences with repeats. In this new version, vsearch looks at all words occurring at least once in the sequences in the initial step. Previously only words occurring exactly once were considered. In addition, vsearch now requires at least 10 words to be shared by the sequences, previously only 6 were required. If the query contains less than 10 words, all words must be present for a match. This change seems to lead to slightly reduced recall, but somewhat increased precision, ending up with slightly improved overall accuracy.

v1.4.6 released October 6th, 2015

Fix subtle bug in SIMD alignment code that reduced accuracy.

v1.4.5 released October 6th, 2015

Fix memory allocation bug when reading long FASTA sequences.

v1.4.4 released October 3rd, 2015

Remove debug message.

v1.4.3 released October 3rd, 2015

Fix a bug with determining amount of memory on some versions of Apple OS X.

v1.4.2 released October 2nd, 2015

Dynamic loading of libraries for reading gzip and bzip2 compressed files if available. Circumvention of missing gzoffset function in zlib 1.2.3 and earlier.

v1.4.1 released September 29th, 2015

Inserted public domain MD5 and SHA1 code to eliminate dependency on crypto and openssl libraries and their licensing issues.

v1.4.0 released September 25th, 2015

Added --sizeorder option.

v1.3.5 released September 17th, 2015

Minor configuration/makefile changes to compile to native CPU and simplify makefile.

v1.3.4 released September 16th, 2015

Fixed compilation problems with zlib and bzip2lib.

v1.3.3 released September 15th, 2015

Fixed bug in hexadecimal digits of MD5 and SHA1 digests. Added --samheader option.

v1.3.2 released September 15th, 2015

Fixed memory leaks. Added -h shortcut for help. Removed extra v in version number.

v1.3.1 released September 14th, 2015

Several new commands and options. Bug fixes.

v1.3.0 released September 9th, 2015

Changed to autotools build system.

v1.2.0–v1.2.19 released July 6th to September 8th, 2015

Several new commands and options added. Bugs fixed. Documentation updated.

v1.1.3 released March 18th, 2015

Fix more bugs with large datasets.

v1.1.2 released March 18th, 2015

Fix bug with large datasets. Fix format of help info.

v1.1.1 released February 20th, 2015

Added info about --log and --quiet options to help text.

v1.1.0 released February 20th, 2015

Added the --quiet option to suppress all output to stdout and stderr except for warnings and fatal errors. Added the --log option to write messages to a log file.

v1.0.16 released February 19th, 2015

Integrated patches from Debian for increased compatibility with various architectures.

v1.0.15 released February 18th, 2015

Fix bug in calculation of identity metric between sequences when using the MBL definition (--iddef 3).

v1.0.14 released February 17th, 2015

Fix a bug where the multiple alignment and consensus sequence computed after clustering ignored the strand of the sequences. Also decreased size of line buffer for reading FASTA files to 1MB again due to excessive stack memory usage.

v1.0.13 released February 17th, 2015

Fix a memory allocation problem when computing multiple sequence alignments with the --msaout and --consout options, as well as a memory leak. Also increased line buffer for reading FASTA files to 4MB.

v1.0.12 released February 6th, 2015

Temporarily fixes a problem with long headers in FASTA files.

v1.0.11 released February 5th, 2015

Introduces the possibility to output results in SAM format (for clustering, pairwise alignment and searching).

v1.0.10 released January 23rd, 2015

Fixes a bug where the --sizein option was ignored and always treated as on, affecting clustering and dereplication commands.

v1.0.9 released January 22nd, 2015

Fixes a bug in the function sorting sequences by decreasing abundance (--sortbysize).

v1.0.8 released January 22nd, 2015

Introduces several changes and bug fixes:

a new linear memory aligner for alignment of sequences longer than 5,000 nucleotides,
a new --cluster_size command that sorts sequences by decreasing abundance before clustering,
meaning of userfields qlo, qhi, tlo, thi changed for compatibility with usearch,
new userfields qilo, qihi, tilo, tihi give alignment coordinates ignoring terminal gaps,
in --uc output files, a perfect alignment is indicated with a = sign,
the option --cluster_fast now sorts sequences by decreasing length, then by decreasing abundance and finally by sequence identifier,
default --maxseqlength value set to 50,000 nucleotides,
fix for bug in alignment in rare cases,
fix for lack of detection of under- or overflow in SIMD aligner.

v1.0.7 released December 19th, 2014

Fixes a bug in the output from chimera detection with the --uchimeout option.

v1.0.6 released December 14th, 2014

Fixes a memory allocation bug in chimera detection (--uchime_ref option).

v1.0.5 released December 9th, 2014

Fixes a minor bug with --allpairs_global and --acceptall options.

v1.0.4 released December 8th, 2014

Added --allpairs_global option. Reduce memory requirements slightly and eliminate memory leaks.

v1.0.3 released December 6th, 2014

Bug fix (now writes help to stdout instead of stderr).

v1.0.2 released December 6th, 2014

Bug fixes (ssse3/sse4.1 requirement, memory leak).

v1.0.1 released December 1st, 2014

Bug fixes (sortbysize, semicolon after size annotation in headers) and minor changes (labels as secondary sort key for most sorts, treat T and U as identical for dereplication, only output size in --dbmatched file if --sizeout specified).

v1.0.0 released November 28th, 2014

First public release.

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.

NAME

DESCRIPTION

VERSION HISTORY

SEE ALSO

CITATION

REPORTING BUGS

AVAILABILITY

COPYRIGHT

GNU General Public License version 3

The BSD 2-Clause License

ACKNOWLEDGMENTS