MMseqs2 Release 5-9375b

Changes since release 4-0b8cc

Bug fixes

• bool flag parameters (e.g. -a) work again
• swapresults will deterministically rank results
• shellcompletion does not report run time anymore

MMseqs2 Release 4-0b8cc

Changes since release 3-be8f6

New features

• Alternative alignments in search (--alt-ali). Find alignments by masking out previously found regions in the target sequence.
• Added map workflow for fast near-exact mapping of reads
• Added easy-linclust workflow, that works on FASTA files
• Sequence lengths longer than 32k are now supported (default sequence length limit is now 65535)
• createdb shuffles the order of entries by default (--dont-shuffle to disable), useful for database splits, where one split could take much longer than others
• linclust now supports MPI
• linclust adds one hash for the whole sequence, to improve extract sequence matching
• New sequence identity computation modes, where the normalization happens on the query or target length instead of alignment length
• New --cov-mode that computes the coverage only based on sequence lengths (--cov-mode 3)
• search/cluster/linclust workflows have learned --alignment-mode 4 for faster ungapped alignments
• Translated search sorts now results by E-value and aggregates all ORFs under the corresponding contig identifier
• prefiltering can now sort hits with score > 255 correctly
• convertalis now works with profiles
• Added generalized database transposition tool swapdb (swapresults only makes sense for prefiltering/alignment results)

Performance

• Speedup extractorf with vectorization
• Many performance improvements to reduce overhead for web server mode
• createtsv writes output in parallel
• Avoid many unnecessary memory allocations in various modules

Bug fixes

• covertmsa does now correctly parses STOCKHOLM files without accession keys
• In search when using splits less than --max-seqs sequences would be the limit, now correctly computes the limit (max-seqs/Splits + 4*sqrt(max_seqs/Splits))
• Fix bug in MsaFilter where wrong sequences would be filtered
• swapresults will add an empty entry if a target entry has no corresponding query match, instead of no entry at all
• createindex creates now correctly creates a tmp directory if no directory exists already
• Fix query split runs for small input databases
• result2stats was reading the wrong first sequence (from query instead of target database)
• result2repseq now writes the correct .dbtype file
• convertalis now reads the correct dbtype for the target sequence
• Fix empty REG_EMPTY bug on macOS
• Fix possible memory corruption when searching against database indexed by 'createindex'
• Report error if -DHAVE_MPI was set and MPI is not installed on the system
• Avoid race condition in kmermatcher (invalid parallel writing to vector)
• Fix msa2profile header output format
• msa2profile uses the FASTA readin mode by default now
• Target profile databases and databases build with --exact-kmer-matching now correctly extract all k-mers
• Fix identical score computation of alignment if clustering using profiles
• Nucleotide backtranslation translateaa would produce invalid codons for X

Others

• removed --early-exit
• Output name of program called

Experimental new modules

• new fast alignment method alignbykmer

Developers

• Cmake flag -DHAVE_GPROF for profiling MMseqs2 using gprof
• Fixed most warnings
• SSTR does not use stringstreams anymore
• Refactored time measuring
• Debug::INFO/WARNING/ERROR is now used consistently across the codebase
• If available (shellcheck)[https://github.com/koalaman/shellcheck] will critique shell scripts and fail the compilation

MMseqs2 Release 3-be8f6

Changes since 2-23394 Release

New Features

• Create simple workflows fasta/fastq in flat file out for clustering easy-cluster and searching easy-search
• Add a new clustering greedy incremental clustering algorithm to the clust module which needs less memory
• Make the new low memory clustering algorithm default if --cov-mode 1 is used in linclust and cluster
• Add alignall module for all-against-all alignments of e.g. clusters
• Improved Windows support
• filterdb learned new modes

Bug fixes

• Fix wrong merging code in linclust
• Fix e-value issues in target-split case
• Fix seg. fault in rescore diagonal if 'z' is used
• Fix seg. fault when using masking in kmermatcher
• Fix wrong filterdb default mode
• prefilter overestimated the required amount of memory and refused to run
• prefilter scores would saturate to early, now they have the full 2^16 range

Others

• Profile searches do create less high scoring false positive through better compositional bias correction and masking of low complexity regions of profiles
• Clustering supports now the whole 2^32 range instead the previously 2^31
• Speed up clustering when using --cov-mode 1
• Rework symlinks to the header databaes
• Support profiles on query and target side in result2profile

MMseqs2 Release 2-23394

Changes since 1-c7a89 Release

New Features

• Translated searches (blastx and tblastn like search modes)
• Improvement splitting input sequences in kmermatcher (Less memory needed for linclust)
• linclust supports nucleotide sequences (experimental feature, k-mer length is not yet optimized)
• search supports nucleotide-nucleotide searches (preview, not stable yet)
• pssm2profile module to print human readable profiles
• msa2profile has a gap match mode to to convert multiple sequences alignments without representative sequence to profile databases
• Compute sequence identity in a similar way to BLAST if --alignment-mode 3 is used
• apply module to execute a arbitrary program on each entry of a mmseqs database. Like map from MapReduce.
• extractorf can use start/stop codons from alternative translation tables
• filterdb now can append entries from other databases by looking them up
• proteinaln2nucl maps a protein alignment back to its original nucleotide sequences
• taxonomy now can blacklist nodes (per default the unclassified and others nodes)
• tmp folder is automatically created, all workflow intermediate results are placed in a subfolder based on the hash of all paths and parameters

Performance Regressions Fixed

• Fixed regression when multiple mmseqs instances were running at the same time

Breaking Command Line Interface Changes

• Incremented index version, old precomputed indices have to be regenerated
• New Profile format, databases generated through convertprofiledb and msa2profile have to be regenerated
• Clustering workflow is now by default cascaded. We replaced the --cascaded flag with --single-step-clustering
• Max sequence length of 32768 is now actually validated and enforced
• Each sequence database has now a dbtype file (AA=0, NUC=1, PROFILE=2)
• extractorf was reworked:
  * --skip-incomplete was split into two parameters --contig-start-mode and --contig-end-mode
  * --longest-orf was reworked into --orf-start-mode
  * removed --extend-min parameter

Others

• Factor four times faster clustering workflow
• Improve speed of linclust by a factor of two
• Remove 'X' from prefilter index (reduces memory and improves speed at the same sensitivity)
• Fix bugs for Query coverage mode (--cov-mode 2)
• Clustering is now the same between single and multi threaded version
• Speedup of kmermatcher
• Fix bug in Clust hash. It can now cluster to 1.0 sequence identity
• Improve target profile search, set max-seqs to infinite for alignments.
• Improve speed of align if prefilter result fit into memory
• Many usability improvements
• Improved suggestions of bash completion
• Expert modules are hidden by default, use -h flag to show everything
• Speed up mergeclusters by a lot
• Fix sequence identity print out bug if the id is less than 10%
• MPI Runner variable can now correctly contain further parameters (RUNNER="mpirun -np 4" was not working)
• Enforcing GCC 4.6 compatibilty in our continous integration

Devlopers

• MMseqs2 can now be included in framework mode to subprojects
• DBReader has a SHUFFLE mode

MMseqs2 Release 1-c7a89

Changes since vNatBiotech Release

New Features

• Taxonomy classification workflow with robust 2bLCA computation and fast LCA computation in O(N LogN)
• Support reading .bz2 archives for createdb
• Createdb can turn multiple fasta files into one database now
• Extend prefilter score range to improve order of best hits after prefiltering.
• Automatically split input sequence set based on system RAM in kmermatcher. Linclust can now run with less memory.

Performance Regressions Fixed

• Fixed underperforming iterative-sequence-profile search without a precomputed index table

Breaking Command Line Interface Changes

• Iterative-non-profile-search --sens-step-size changed to --sens-steps (Number of Iterations) (Does not break nested workflows anymore)

Others

• Query coverage mode (--cov-mode 2) for searching
• Clustering is now the same between single and multi threaded version
• Bug fixes in rescorediagonal
• Speedup of kmermatcher
• Speedup and memory reduction of swapresults
• Many usability improvements

Devlopers

• MMseqs2 can now be included in framework mode to subprojects

Nature Biotechnology Release

Release for Nature Biotechnology