Skip to content

Latest commit

 

History

History
49 lines (43 loc) · 4.98 KB

CHANGELOG.md

File metadata and controls

49 lines (43 loc) · 4.98 KB
Raw

Changelog - Pancake

Active version in development

Changes

v1.2.0

Version

  • pancake - 2f05f3e16219291485de130f7ca68dc40ae7c5c5 (Mar 11, 2021), pancake 1.2.0 (commit SL-release-10.0.0-417-g2f05f3e)

Changes

  • The AlignmentResult now stores the number of alignment differences, so that the client doesn't have to parse the CIGAR string again. Updated the aligners to produce the diff counts.
  • FastaSequenceCachedStore implementation. This is a store for FastaSequenceCached objects, which are like a "view" into the FASTA sequences (they do not own the data, only point to it). This is now used in the SeqDBReaderCachedBlock.
  • Generic interface to MapperHiFi. It's now possible to run a single function which takes a set of std::string targets and queries, and maps/aligns them.
  • Added the SeqDB dump tool (seqdb-dump) to dump the entire DB or just one block.
  • Added the SeqDB info tool (seqdb-info) to display information about the SeqDB (length, N50, etc.).
  • Minor refactoring, versioning code.
  • (non-IPA related) KSW2 third-party library now removed from Pancake, and used from Pbcopper.
  • (non-IPA related) Refactoring the MapperCLR: parametrized the flankExtensionFactor setting ; and always computing alignment regions during mapping, instead of only when performing alignment.
  • (non-IPA related) AlignerBatchCPU implementation - an aligner which first takes a batch of sequence pairs, then performs alignment of all submitted pairs in parallel, on CPU.
  • (non-IPA related) MapperBatchCPU implementation - a batch mapper which first takes a batch of sequence pairs, then performs mapping of sequences in parallel on the CPU. Following that, the mappings are aligned in batch using the AlignerBatchCPU in parallel on the CPU.
  • (non-IPA related) AlignerBatchGPU - analogous to the AlignerBatchGPU, but uses the Cudaaligner (GenomeWorks) to perform alignment of the sequence pairs on the GPU.
  • (non-IPA related) MapperBatchGPU - analogous to the MapperBatchCPU, but this uses the GPU aligner to align sequences. Mapping is still performed on the CPU, same as with the MapperBatchCPU.
  • (non-IPA related) Abstract class MapperBatchBase added, and MapperBatchCPU and MapperBatchGPU now both inherit the interface.
  • (non-IPA related) Optional dependency injection of the FireAndForget object into the AlignerBatchCPU, MapperBatchCPU and MapperBatchGPU to better handle kernel overload.
  • All CLR mappers (MapperCLR, MapperBatchCPU, MapperBatch GPU) can now optionally skip/mock perfect alignments, and skip symmetric overlaps. (Mocking means that if a self-hit is detected based on sequence IDs, a perfect alignment result will be generated without actually computing the alignment.)
  • Fixed the bestNSecondary in MapperCLR::Map_ where it did not properly respect the value < 0. (Instead of reporting all alignments, only some were reported.)
  • The batch mappers (MapperBatchCPU, MapperBatchGPU) can accept custom mapping parameters for every chunk in a batch.
  • Converted the unique_ptr to references in the overlap writers.
  • Fixed the score computation in MapperCLR.

v1.1.0 - SL - Release 10.1.0

Version

  • pancake - commit 29844f96d5cf58874a04ffd6a9895abe00a0750b (origin/release/prep) (Dec 16, 2020), pancake 1.1.0 (commit SL-release-10.0.0-186-g29844f9)

Changes

  • LIS chaining of seed hits. Improvements in overlap sensitivity by refining seed hits before alignment. The seed hits are now refined using the Longest Increasing Subsequence (LIS) algorithm.
  • Scalability improvement in Pancake - reduced system time by removing std::clock() usage. It turns out that this function is blocking.
  • Filtering duplicate overlaps in Pancake, which can occur when there is a larger gap in between seed hits.
  • Pbcopper subproject now points to the GitHub mirror.
  • GitHub mirror of Pancake now exists: https://github.com/PacificBiosciences/pancake
  • Handling edge cases in alignment explicitly (e.g. when either query or target is of length 0). Some aligners behaved non-deterministically in these cases.
  • Resolved warnings.
  • Minor refactoring.
  • Encoded seed span. Seed span is now encoded in the Seed structure. The same as in Minimap2. Span of a seed hit in both query and target is now encoded in the SeedHit structure. Seed indexing and hit collection now works correctly in the homopolymer-compressed mode. It also works with the spaced seeds feature. The new maximum kmer size is equal to 28 now. This is because the span is encoded within the seed data structure, so some bits had to be specialized. A lot of tests had to be updated because previously they used the default of k = 30.
  • (non-IPA related) Added MapperCLR - implemented a CLR mapper, currently available via API. and several more followups to resolve warnings and improve API

v0.2.0 - First release - SL - Release 10.0.0 (SL-release-10.0.0)

Version

  • pancake - commit dd98868876ff0171be3e191973834e1ac5229123 (HEAD, tag: SL-release-10.0.0, origin/release/10.0.0) (Sep 23, 2020) pancake 0.2.0 (commit SL-release-10.0.0)