Changelog

Changelog#

[0.4.0a3] - ****--#

Features

  • Document that the zarr-vcf dataset can be either a path or an in-memory zarr group. (feature introduced in #966, documented in #974, @hyanwong)

Fixes

  • Properly account for “N” as an unknown ancestral state, and ban “” from being set as an ancestral state (#963, @hyanwong)

[0.4.0a2] - 2024-09-06#

2nd Alpha release of tsinfer 0.4.0

Features

  • Add batch ancestor and sample matching APIs for splitting work across many independent jobs. (#954, #917, @benjeffery)

[0.4.0a1] - 2024-07-26#

Alpha release of tsinfer 0.4.0

Features

  • tsinfer now supports inferring data from an vcf-zarr dataset. This allows users to infer from VCFs via the optimised and parallel VCF parsing in bio2zarr.

  • The VariantData class can be used to load the vcf-data and be used for inference.

  • vcf-zarr sample_ids are inserted into individual metadata as variant_data_sample_id if this key does not already exist.

Breaking Changes

  • Remove the uuid field from SampleData. SampleData equality is now purely based on data. (#748, @benjeffery)

Performance improvements

  • Reduce memory usage when running match_samples against large cohorts containing sequences with substantial amounts of error. (#761, @jeromekelleher)

  • truncate_ancestors no longer requires loading all the ancestors into RAM. (#811, @benjeffery)

  • Reduce memory requirements of the generate_ancestors function by providing the genotype_encoding (#809) and mmap_temp_dir (#808) options (@jeromekelleher).

  • Increase parallelisation of match_ancestors by generating parallel groups from their implied dependency graph. (#828, #147, @benjeffery)

[0.3.3] - 2024-07-17#

Fixes

  • Bug fix release for numpy 2 (#937).

Breaking Changes

[0.3.2] - 2024-07-16#

Features

  • tsinfer now supports numpy2 (and 1.XX) and python 3.12.

Breaking Changes

  • tsinfer now requires Python 3.9 or later

[0.3.1] - 2023-04-19#

  • Bug fix release for a bad dependency specification.

[0.3.0] - 2022-10-25#

Features

  • When calling sample_data.add_site() the ancestral state does not need to be the first allele (index 0): alternatively, an ancestral allele index can be given (and if MISSING_DATA, the ancestral state will be imputed). (#718, #686 @hyanwong)

  • The CLI interface now allows recombination rate (or rate maps) and mismatch ratios to be specified (#731, #435 @hyanwong)

  • The calls to match-ancestors and match-samples via the CLI are now logged in the provenance entries of the output tree sequence (#732 and 741, #730 @hyanwong)

  • The CLI interface allows --no-post-process to be specified (for details of post- processing, see “Breaking changes” below) (#727, #721 @hyanwong)

  • matching routines warn if no inference sites (#685, #683 @hyanwong)

Fixes

  • sample_data.subset() now accepts a sequence_length (#681, @hyanwong)

  • verify no longer raises error when comparing a genotype to missingness. (#716, #625, @benjeffery)

Breaking changes:

  • The simplify parameter is now deprecated in favour of post_process, which prior to simplification, removes the “virtual-root-like” ancestor (inserted by tsinfer to aid the matching process) then splits the ultimate ancestor into separate pieces. If splitting is not required, the post_process step can also be called as a separate function with the parameter split_ultimate=False (#687, #750, #673, @hyanwong)

  • Post-processing by default erases tree topology that exists before the first site and one unit after the last site, to avoid extrapolating into regions with no data. This can be disabled by calling post_process step as a separate function with the parameter erase_flanks=False (#720, #483, @hyanwong)

  • Inference now sets time_units on both ancestor and final tree sequences to tskit.TIME_UNITS_UNCALIBRATED, stopping accidental use of branch length calculations on the ts. (#680, @hyanwong)

[0.2.3] - 2022-04-08#

Features

Fixes

  • Mark zarr 2.11.0, 2.11.1 and 2.11.2 as incompatible due to zarr-python bugs #965 and #967. (#643, #657, @benjeffery)

[0.2.2] - 2022-02-23#

Bugfixes:

  • Mutations at non-inference sites are now guaranteed to be fully parsimonious. Previous versions required a mutation above the root when the input ancestral state disagreed with the ancestral state produced by the parsimony algorithm. Now fixed by using the new map_mutations code from tskit 0.3.7 (#557, @hyanwong)

New Features:

Breaking changes:

  • Oldest nodes in a standard inferred tree sequence are no longer set to frequencies ~2 and ~3 (i.e. 2 or 3 times as old as all the other nodes), but are spaced above the others by the mean time between unique ancestor ages (#485, @hyanwong)

  • The tsinfer.SampleData.from_tree_sequence() function now defaults to setting use_sites_time and use_individuals_time to False rather than True (#599, @hyanwong)

[0.2.1] - 2021-05-26#

Bugfix release

Bugfixes:

  • Fix a bug in the core LS matching algorithm in which the rate of recombination was being incorrectly computed (#493, #514, @jeromekelleher, @hyanwong).

  • tsinfer.verify() no longer requires that non-ancestral alleles in a SampleData and Tree Sequence file are in the same order (#490, #492, @hyanwong).

New Features:

  • Inferred ancestral haplotypes may be truncated via AncestorData.truncate_ancestors() to improve performance when inferring large datasets (#276, #467, @awohns).

Breaking changes:

  • tsinfer now requires Python 3.7

[0.2.0] - 2020-12-18#

Major feature release, including some incompatible file format and API updates.

New features:

  • Mismatch and recombination parameters can now be specified via the recombination_rate and mismatch_ratio arguments in the Python API.

  • Missing data can be accomodated in SampleData using the tskit.MISSING_DATA value in input genotypes. Missing data will be imputed in the output tree sequence.

  • Metadata schemas for population, individual, site and tree sequence metadata can now we be specified in the SampleData format. These will be included in the final tree sequence and allow for automatic decoding of JSON metadata.

  • Map non-inference sites onto the tree by using the tskit map_mutations parsimony method. This allows us to support sites with > 2 alleles.

  • Historical (non-contemporaneous) samples can now be accommodated in inference, assuming that the true dates of ancestors have been set, by using the concept of “proxy samples”. This is done via the new function AncestorData.insert_proxy_samples(), then setting the new parameter force_sample_times=True when matching samples.

  • The default tree sequence returned after inference when simplify=True retains unary nodes (i.e. simplify is done with keep_unary=True.

Breaking changes:

  • The ancestors tree sequence now contains the real alleles and not 0/1 values as before.

  • Times for undated sites now use frequencies (0..1), not as counts (1..num_samples), and are now stored as tskit.UNKNOWN_TIME, then calculated on the fly in the variants() iterator.

  • The SampleData file no longer accepts the inference argument to add_site. This functionality has been replaced by the exclude_positions argument to the infer and generate_ancestors functions.

  • The SampleData format is now at version 5, and older versions cannot be read. Users should rerun their data ingest pipelines.

  • Users can specify variant ages, via sample_data.add_sites(... , time=user_time). If not None, this overrides the default time position of an ancestor, otherwise ancestors are ordered in time by using the frequency of the derived variant (#143).

  • Change “age” to “time” to match tskit/msprime notation, and to avoid confusion with the age since birth of an individual (#149). Together with the 2 changes below, this addition bumped the file format to 3.0.

  • Add the ability to record user-specified times for individuals, and therefore the samples contained in them (currently ignored during inference). Times are added using sample_data.add_individual(... , time=user_time) (#190).

  • Change tsinfer.UNKNOWN_ALLELE to tskit.MISSING_DATA for marking unknown regions of ancestral haplotypes (#188) . This also involves changing the allele storage to a signed int from np.uint8 which matches the tskit v0.2 format for allele storage (see https://github.com/tskit-dev/tskit/issues/144).

Bugfixes:

  • Individuals and populations in the SampleData file are kept in the returned tree sequence, even if they are not referenced by any sample. The individual and population ids are therefore guaranteed to stay the same between the sample data file and the inferred tree sequence. (#348)

[0.1.4] - 2018-12-12#

Bugfix release.

  • Fix issue caused by upstream changes in numcodecs (#136).

[0.1.3] - 2018-11-02#

Release corresponding to code used in the preprint.

[0.1.2] - 2018-06-18#

Minor update to take advantage of msprime 0.6.0’s Population and Individual objects and fix various bugs.

Breaking changes:

  • Bumped SampleData file format version to 1.0 because of the addition of individuals and populations. Older SampleData files will not be readable and must be regenerated.

  • Changed the order of the alleles and genotypes arguments to SampleData.add_site.

New features:

  • Sample and individual metadata now handled correctly.

  • Added –no-simplify option to CLI and simplify=True option to infer function.

  • Better handling of missing files (raises correct exceptions).

  • tsinfer list now presents basic information for .trees files.

Bug fixes:

  • Degenerate examples with zero inference sites are now rooted (#44)

  • Simplify=False results in tree sequence with correct sample nodes.