v1.3.5

New querying options

• -Q, --files-as-queries: It is now possible to use a file containing one or multiple sequences as one query.
• -E, --min-nb-colors: Requires that a query occurs in a user-provided minimum number of colors to report the query as present in the graph

Bug fixes

• Fix bug with incorrect read positions in input files causing suboptimal construction time
• Fix bug with the resizing of MinimizerIndex and KmerHashtable causing program hanging for very small input data

Refactor

• CLI updated. Querying works a little differently from previous versions. Default behavior for previous versions was to report presence/absence of each query. New behavior is to report the number of k-mers from each query shared with the graph or with its colors, unless parameter -e is used explicitly in which case presence/absence is reported. Parameter -p will make the querying report a ratio of found k-mers (with respect to the number of k-mers in each query) rather than a number
• README updated: New "Benchmarking" section provides guidelines on how to properly compare other tools to Bifrost, use cases updated, citation updated, installation description simplified, etc.

v1.3.1

Minor upgrade to 1.3.0:

• Fix compilation issues for MacOS and FreeBSD
• Fix issue with reporting the ratio of k-mers found in each query for colored dBGs
• Fix issue with the inexact k-mer search in colored dBGs

v1.3.0

Bifrost memory usage and running time has improved by up to 30% (depending on the input dataset)

• The minimizer index and kmer hash-table use now robin-hood hashing instead of linear probing. They default to 95% maximum bucket occupancy instead of 80% in previous versions. Furthermore, hash table resizing uses now a 20% memory increase instead of doubling the size.
• Blocked Bloom Filter uses a default 24 bits per key to enable a much faster joining step.
• Extracting the exact unitigs from the Blocked Bloom Filter is done in 3 steps instead of 2: approximate unitigs are extracted to disk, read from disk and indexed in memory before annotating k-mers with their respective counts. Reader-writer locks can be avoided this way. However, Bifrost now uses a bit of disk space.
• Code refactoring

New options for querying

• -p, --nb-found-km: Output the number of found k-mers for each query (disable parameter -e)
• -P, --ratio-found-km: Output the ratio of found k-mers for each query (disable parameter -e)
• Fix rounding issue in #67

This version breaks Bifrost Index File (.bfi) compatibility with previous Bifrost versions, i.e. v1.3.0 cannot read .bfi files generated by v1.2.1 and older versions.

v1.2.1

Update kseq to latest version:

• Faster reading from FASTA/FASTQ
• Reads sequences larger than 2 GB
• Experiments show lower memory usage

v1.2.0

• GFA and FASTA output are now compressed by default. Compressed GFA/FASTA can be read by Bifrost.
• Optional Bifrost binary output (instead of GFA/FASTA). Uses less disk space than GFA/FASTA and loads faster in memory.
• Bifrost index output by default. Enables to load a Bifrost graph (GFA/FASTA or binary) much faster in memory.
• Time and memory reduction on some datasets.
• Minor bug fixes.

API and CLI compatibility breaks with previous versions but only minor changes (see Changelog). GCC version minimum requirement changes to 5.1 (from 4.8).

v1.0.6.5

• Major bugfix for single-threaded construction
• Minor bugfixes

v1.0.6.4

Minor update to 1.0.6.3:

• Remove XXHash dependency from install which had been put back by mistake in the cmake files.

v1.0.6.3

Minor update to v1.0.6.2:

• Add MAX_GMER_SIZE compile variable to choose the maximum size of g-mers (minimizers). Works exactly the same as MAX_KMER_SIZE. By default, MAX_GMER_SIZE is equal to MAX_KMER_SIZE.
• Fix an issue with the update function where the k-mer size of the input graph was not used to make the output graph (default k-mer size was used instead).

v1.0.6.2

Overall, Bifrost is now much faster and a lot more memory efficient compared to v1.0.5.

• Improvement on the rolling hash-function RepHash (much less collisions for small values).
• Blocked Bloom Filter cannot use over-loaded blocks anymore. When using the 2-blocks hashing method, if the 2 blocks are at >65% capacity, the minimizer is rehashed to find 2 new blocks which are hopefully not over-loaded. After 8 iterations of 2-overloaded blocks, the k-mer hash is stored in an unsorted_set.
• AVX2 version of the Blocked Bloom Filter is disabled at the moment. It doesn't bring any performance gain over the non-AVX2 version anymore. Need some rework.
• XXHash replaced with Wyhash

v1.0.5

Fix major problem with unitig deletion in colored dBGs