Medaka

Minor fix release

Fixed
- Fix a memory error in pileup calculation.
Added
- Update variant calling models and benchmarks.

Minor fix release

Fixed
- Detect NaNs during training and halt early.
- Workaround pysam interface changes (for conda package).
Added
- Preliminary hard-RLE model for R9.4.1
- --regions argument can now be a .bed file.
- Support soft-RLE network training.

This release includes an experimental consensus mode using run-length encoded
alignments. Use of this algorithm can be specified using the new "rle" model:

medaka_consensus -m r941_min_high_g340_rle -i basecalls.fasta -d draft.fa

Feature release

Added
- Consensus models for guppy 3.3 and 3.4.
- Aggregation of Guppy modified base probability tables.
- Multi-thread stitching of inference chunks in `medaka_consensus`.
- Optionally run whatshap phase at the end of `medaka_variant`.

Minor fix release

Fixed
- Fix bug where feature matrix was misaligned with coordinate system.
- Fixed issue with `medaka_variant` failing on zero-coverage regions.
- Rename incorrectly named diploid SNP calling model.
- Made variant calling faster by resolving trivial bottleneck in variant classification.
Added
- Add missing arguments from `smolecule` command.
- Output contig names are no longer written as samtools-style regions.

Feature release

Fixed
- Corrected parsing of region strings with multiple `:` charaters
- Fixed bug causing larger than requested overlap in inference chunks.
- Fixed rare consensus stitching error.
Added
- Added a `-f` force overwrite option to `medaka_consenses`.
- Added *C. elegans* assembly benchmarks to documentation.
Changed
- Switched variant calling to an explicitely diploid calling model.
- Refreshed *E. coli* benchmark to include effect of `racon`.
- Refreshed variant calling benchmarks.

Minor fix release.

Fixed
- Additional fix to handling lowercase reference sequences.
- Fix bug in creation of RLE alignments.
Added
- Update `update_model.py` script.
Changed
- Unify how LabelSchemes store training data.
Removed
- Remove option to select labelling scheme during training.