# Change Log
All notable changes to this project will be documented in this file.
This project adheres to [Semantic Versioning](http://semver.org/).
## [2.4.0] - 2020-11-05
- Allow scenarios to contain samples for which no BAM files are available. This allows to e.g. model tumor/normal from just the tumor sample with known contamination. Resulting probabilities will properly reflect the uncertainty about whether a variant is somatic or germline.
- Speed up SNV and MNV computations by precomputed call and miscall likelihoods.
- Add a flag --pairhmm-mode [exact|fast], that allows to instruct varlociraptor to only compute the optimal path in the pairHMM. This should be much faster in practice, but can come with some wrong likelihoods in rare extreme cases. Advice: only use on large cohorts or where exact allele frequencies do not matter.
- Fix insert size handling on single end samples (@dlaehnemann).
## [2.3.0] - 2020-09-09
- Include read orientation bias into the model.
- Exlcude softclipped and non-standard orientation reads from SNV and MNV calling as they are indicative of SVs and often cause artifact substitutions while sometimes not being reflected via a higher uncertainty in the MAPQ. Not considering them is the conservative choice.
## [2.2.1] - 2020-08-24
- Allow to set reference buffer size (--reference-buffer-size), for improved parallelization when calling SVs.
- Fix breakend handling getting confused between events when calling BCF with multiple breakends and more than a single thread.
## [2.2.0] - 2020-08-24
- Allow parallelization via setting the number of threads.
- BCF output is now unsorted, and must be sorted afterwards with bcftools.
- Imprecise variants are skipped for now, until proper support is implemented.
## [2.1.0] - 2020-08-11
- Infer missing antisense breakends (sometimes callers only report one direction although the other is necessary as well).
- Support for single breakends.
- Support for arbitrary replacements.
- Fixed a bug with posterior odds filtration leading to underestimation of the odds.
## [2.0.1] - 2020-08-06
- Fixed allele frequency biases that occurred due to missed evidence when investigating breakends.
- Fixed pattern-too-long error that occurred in some corner cases.
- Fixed error occurring when investigation breakends without event tag. These are skipped for now. Special handling will be added later.
## [2.0.0] - 2020-07-09
- Add support for inversions, deletions and breakends.
- Slightly modified CLI options.
## [1.7.3] - 2020-06-17
- Polished TMB plots.
- Allow to omit insert size evidence. This is important for amplicon data, where indels do not impact insert size.
## [1.7.2] - 2020-06-03
- Fixed possible values in CLI.
## [1.7.1] - 2020-06-02
- Fixed compiler issue when building varlociraptor with certain rust versions.
- Use latest rust-bio release.
## [1.7.0] - 2020-05-27
- Improved TMB estimation plots, now offering three modes (hist, curve, and stratified), as well as being able to better see multiple scales.
## [1.6.4] - 2020-03-18
- Report negative SVLEN for deletions again. The htslib bug is actually fixed already. However, it is mandatory to rerun varlociraptor preprocess to avoid it downstream.
## [1.6.3] - 2020-03-17
- Cleanup of debug messages.
## [1.6.2] - 2020-03-17
- Fix odds filtering arguments.
- Improved error messages.
- Improved insert size estimation.
- Use htslib 1.10.
- Work around htslib bug that misinterprets negative SVLENs: SVLEN is now always positive.
## [1.5.0] - 2019-12-04
- Introduce SNV selectors in the event formula language.
- Work around a segmentation fault in Htslib when estimating tmb.
## [1.4.4] - 2019-12-03
- Modify observation format such that encoding cannot lead to corner cases where BCF interprets a value as a vector end marker.
- Fix test case generation
## [1.4.3] - 2019-12-02
- Make observation reading and writing more robust by controlling the length of INFO field vectors. This fixes a bug when reading leads to invalid strand information.
## [1.4.2] - 2019-12-01
- Avoid unsafe memory operations when writing and reading preprocessed observations.
## [1.4.1] - 2019-11-29
- Work around a segmentation fault caused by Htslib.
- Update command line usage instructions.
## [1.4.0] - 2019-11-28
- Separate calling and preprocessing of observations. This allows to easily reuse large parts of the computation when changing the scenario. Further, it allows to parallelize across samples.
## [1.3.0] - 2019-11-13
- Enable setting of the strandedness of the sequencing protocol at the command line (same or opposite).
- Allow contig-specific definition of the allele frequency universe in the grammar.
## [1.2.2] - 2019-10-04
- Explicitly annotate unit (PHRED, or linear) of probabilities in BCF output.
## [1.2.1] - 2019-09-27
- Fixed a bug in usage of htslib that could lead to a segmentation fault when estimating TMB.
## [1.2.0] - 2019-09-27
- Estimation of tumor mutational burden as a curve of expectations over minimum allele frequencies.
- Various small bug fixes.
## [1.1.0] - 2019-07-03
- Numerical improvements to strand bias model.
## [1.0.1] - 2019-05-15
- Fix numerical issue with indel variant calling.
- Fix indel window length issues with reads >100bp.
## [1.0.0] - 2019-05-09
- Rewrite of statistical model, using rust-bios new trait machinery.
- Generalization of the model, allowing arbitrary allele frequency events and sample numbers.
- Inclusion of strand bias in the model.
- Better VCF output.
- Various bug fixes.
- New test framework.
- CLI overhaul.
- Experimental tumor-normal CNV calling.
- Various improvements to calling accuracy and fixed corner cases.
## [0.7.0] - 2018-11-06
### Changed
- overhaul of FDR machinery to make it only one tool that outputs a BCF filtered at a provided FDR threshold
- a number of performance optimizations leading to more than an order of magnitude speedup of ProSolo, a tool using this library, [with details described in its repo](https://github.com/ProSolo/prosolo/issues/2); most importantly these changes are:
- [caching the CIGAR string of reads](https://github.com/varlociraptor/varlociraptor/pull/34)
- caching likelihood point estimates for reuse in different (two-dimensional) Events (see PRs [36](https://github.com/varlociraptor/varlociraptor/pull/36) and [40](https://github.com/varlociraptor/varlociraptor/pull/40))
- [caching of `prob_rho()` function in the SingleCellBulkModel](https://github.com/varlociraptor/varlociraptor/pull/46)
- [minimize the number of `ln` operations performed, caching some values per program run, per pileup or per observation](https://github.com/varlociraptor/varlociraptor/pull/48)
- remove use of flamegraphing crates in favor of [perf flamegraphing](https://gist.github.com/dlaehnemann/df31787c41bd50c0fe223df07cf6eb89)
- [minor bugfix for `ContinuousAlleleFreqs` ranges depicting points](https://github.com/varlociraptor/varlociraptor/pull/47)
- introduced `cargo fmt` use, including in continuous integration tests
- [insert size estimation from data and handling as `alignment_properties`, removing the necessity to provide them for each run via command-line arguments](https://github.com/varlociraptor/varlociraptor/pull/41)
- [binarise MAPQ to 0 vs. maximum observed MAPQ at a site, to remove reference mapping bias](https://github.com/varlociraptor/varlociraptor/pull/38) -- using the given instead of the binarised MAPQ could be selected via a command line option in the downstream tool that sets `use_mapq`
- overhaul of observation extraction mechanism and sampling probability calculation for indels, looking at full fragments (e.g. read pairs) jointly instead of looking at reads separately( see PRs [28](https://github.com/varlociraptor/varlociraptor/pull/28), [29](https://github.com/varlociraptor/varlociraptor/pull/29) and [30](https://github.com/varlociraptor/varlociraptor/pull/30))
- [added functionality to calculate likelihoods and calls for candidate sites without a candidate alternative allele, i.e. homozygous reference candidates](https://github.com/varlociraptor/varlociraptor/pull/24)
- dependency updates (`rust-htslib` to `0.22`, `rust-bio` to `0.23`)
## [0.6.0] - 2018-01-12
### Changed
- added functionality to filter by FDR thresholds based on sets of `Events` and improved code for summing up Event likelihoods (incl. fixing numerical overshoot issues)
- dependency updates (`rust-htslib` to `0.16`, `rust-bio` to `0.16` from `rustc-serialize` to `csv` and `serde`)
- some test fixes
## [0.5.0] - 2017-11-17
### Changed
- FDR control now works for sets of events.
- The pair HMM implementation has been finetuned and improved.
- Artifacts from weird mapper decisions are now better detected.
## [0.4.0] - 2017-07-06
### Changed
- Refactored and fixed the false discovery rate estimation code.
- Improved indel likelihood calculation (pairHMM based indel likelihoods; do not normalize indel likelihoods for ref and alt, only consider a small window around the indel)
- New SingleCellBulkModel for calling variants in single cell data against a bulk background sample.
- Update rust-bio dependency to 0.14.* or newer.
- Marginalization only over specified Events, not the full allele frequency ranges.
- Full-blown Cigar string parsing with CigarErrors and full single nucleotide variant (SNV) support, externalized via rust-htslib dependency.
- New Ranges with inclusive/exclusive end points.
- Added commitizen.
## [0.3.0] - 2017-05-04
### Changed
- Ensure fair sampling of reads supporting ALT and REF allele.
## [0.2.0] - 2017-03-28
### Changed
- calculate ALT and REF likelihood by realigning against pseudo-haplotypes
## [0.1.0] - 2016-11-02
### Added
- initial release