Changelog#

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

v0.19.0#

Additions#

  • Add a mask_obs argument to tl.clonotype_network that allows to compute the clonotype networks on a subset of the cells (#557).

  • Add datasets.stephenson2021_5k, an example dataset for the upcoming BCR tutorial (#565)

Fixes#

  • Add all optional dependencies required for testing to the [test] dependency group (#562).

  • Unpin AnnData version (#551)

v0.18.0#

Additions#

  • Isotypically included B cells are now labelled as receptor_subtype="IGH+IGK/L" instead of ambiguous in tl.chain_qc (#537).

  • Added the normalized_hamming metric to pp.ir_dist that accounts for differences in CDR3 sequence length (#512).

  • tl.define_clonotype_clusters now has an option to require J genes to match (same_j_gene=True) in addition to same_v_gene. (#470).

Performance improvements#

  • The hamming distance has been reimplemented with numba, achieving a significant speedup (#512).

  • Clonotype clustering has been accelerated leveraging sparse matrix operations (#470).

Fixes#

  • Fix that pl.clonotype_network couldn’t use non-standard obsm key (#545).

Other changes#

  • Make parasail an optional dependency since it is hard to install it on ARM CPUs. TCRdist is now the recommended default distance metric which is much faster than parasail-based pairwise sequence alignments while providing very similar results (#547).

  • Drop support for Python 3.9 in accordance with SPEC0 (#546)

v0.17.2#

Fixes#

  • Detection of CPU count in define_clonotype_clusters was broken (#527)

v0.17.1#

Fixes#

  • Compatibility with numpy 2.0 (#525)

Chore#

  • scverse template update to v0.4 (#519)

v0.17.0#

Additions#

  • Add “TCRdist” as new metric (#502)

Fixes#

  • Fix issue with detecting the number of available CPUs on MacOS (#518)

v0.16.1#

Fixes#

  • Fix default value for n_jobs in ir.tl.ir_query that could lead to an error (#498).

  • Update description of D50 diversity metric in documentation (#499).

  • Fix clonotype_modularity not being able to store result in MuData in some cases (#504).

  • Fix issue with creating sparse matrices from generators with the latest scipy version (#504)

v0.16.0#

Backwards-incompatible changes#

  • Use the umi_count field instead of duplicate_count to store UMI counts. The field umi_count has been added to the AIRR Rearrangement standard in version 1.4 (#487). Use of duplicate_count for UMI counts is now discouraged. Scirpy will use umi_count in all scirpy.io functions. It will not change AIRR data that is read through scirpy.io.read_airr that still uses the duplicate_count column. Scirpy remains compatible with datasets that still use duplicate_count. You can update your dataset using

    adata.obsm["airr"]["umi_count"] = adata.obsm["airr"]["duplicate_count"]
    

Other#

  • the io.to_dandelion and io.from_dandelion interoperability functions now rely on the implementation provided by Dandelion itself (#483).

v0.15.0#

Fixes#

  • Fix incompatibility with scipy 1.12 (#484)

  • Fix incompatibility with adjustText 1.0 (#477)

  • Reduce overall importtime by deferring the import of the airr package until it is actually used. (#473)

New features#

  • Speed up alignment distances by pre-filtering. There are two filtering strategies: A (lossless) length-based filter and a heuristic based on the expected penalty per mismatch. This is implemented in the FastAlignmentDistanceCalculator class which supersedes the AlignmentDistanceCalculator class, which is now deprecated. Using the "alignment" metric in pp.ir_dist now uses the FastAlignmentDistanceCalculator with only the lenght-based filter activated. Using the "fastalignment" activates the heuristic, which is significantly faster, but results in some false-negatives. (#456)

  • Switch to joblib/loky as a backend for parallel processing in pp.ir_dist. Joblib enables to switch to alternative backends that support out-of-machine computing (e.g. dask, ray) via the parallel_config context manager. Additionally, chunk sizes are now adjusted dynamically based on the problem size. (#473)

Documentation#

  • The default values of the distance calculator classes in ir_dist.metrics was unclear. The default value is now set in the classes. In pp.ir_dist and ir_dist.sequence_dist, no cutoff argument is passed to the metrics objects, unless one is explicitly specified (previously None was passed by default).

v0.14.0#

Breaking changes#

  • Reimplement pp.index_chains using numba and awkward array functions, achieving a significant speedup. This function behaves exactly like the previous version except that callback functions passed to the filter arguments must now be vectorized over an awkward array, e.g. to check if a junction_aa field is present you could previously pass lambda x: x['junction_aa'] is not None, now an accepted version would be lambda x: ~ak.is_none(x["junction_aa"], axis=-1). To learn more about native awkward array functions, please refer to the awkward array documentation. (#444)

Additions#

  • The clonal_expansion function now supports a breakpoints argument for more flexible “expansion categories”. The breakpoints argument supersedes the clip_at parameter, which is now deprecated. (#439)

Fixes#

  • Fix that define_clonotype_clusters could not retreive within_group columns from MuData (#459)

  • Fix that AIRR Rearrangment fields of integer types could not be written when their value was None (#465)

v0.13.1#

Fixes#

  • Fix that clonotype_modularity could not run with AnnData object (#421).

  • Fix usage of wrong column in 3k tutorial (#423)

  • Change igraph dependency from python-igraph to igraph (#436)

  • Fix that group_abundance didn’t work when AIRR data was stored in a different MuData slot than airr (#438)

v0.13.0 - new data structure based on awkward arrays#

This update introduces a new datastructure based on awkward arrays. The new datastructure is described in more detail in the documentation and is considered the “official” way of representing AIRR data for scverse core and ecosystem packages.

Benefits of the new data structure include:

  • a more natural, lossless representation of AIRR Rearrangement data

  • separation of AIRR data and the receptor model, thereby getting rid of previous limitations (e.g. “only productive chains”) and enabling other use-cases (e.g. spatial AIRR data) in the future.

  • clean adata.obs as AIRR data is not expanded into columns

  • support for MuData for working with paired gene expression and AIRR data as separate modalities.

The overall workflow stays the same, however this update required several backwards-incompatible changes which are summarized below.

Backwards-incompatible changes#

New data structure#

Closes issue https://github.com/scverse/scirpy/issues/327.

Changed behavior:

  • there are no “has_ir” and “multichain” columns in adata.obs anymore

  • By default all fields are imported from AIRR rearrangement and 10x data.

  • The restriction that all chains added to an AirrCell must have the same fields has been removed. Missing fields are automatically filled with missing values.

  • io.upgrade_schema can update from v0.7 to v0.13 schema. AnnData objects generated with scirpy <= 0.6.x cannot be read anymore.

  • pl.spectratype now has a chain attributed and the meaning of the cdr3_col attribute has changed.

New functions:

  • pp.index_chains

  • pp.merge_chains

Removed functions:

  • pp.merge_with_ir

  • pp.merge_airr_chains

API supporting MuData#

Closes issue https://github.com/scverse/scirpy/issues/383

All functions take (where applicable) the additional, optional keyword arguments

  • airr_mod: the modality in MuData that contains AIRR information (default: “airr”)

  • airr_key: the slot in adata.obsm that contains AIRR rearrangement data (default: “airr”)

  • chain_idx_key: the slot in adata.obsm that contains indices specifying which chains in adata.obsm[airr_key] are the primary/secondary chains etc.

New class:

  • util.DataHandler

Updated example datasets#

The example datasets have been updated to be based on the new datastructure and are now based on MuData.

  • The example datasets have been regenerated from scratch using the loader notebooks described in the docstring. The Maynard dataset gene expression is now based on values generated with Salmon instead of RSEM/featurecounts.

  • Scirpy now uses pooch to manage example datasets.

Cleanup#

  • Removed the deprecated functions io.from_tcr_objs, io.from_ir_objs, io.to_ir_objs, pp.merge_with_tcr, pp.tcr_neighbors, pp.ir_neighbors, tl.chain_pairing

  • Removed the deprecated classes TcrCell, AirrChain, TcrChain

  • Removed the function pl.cdr_convergence which was never public anyway.

Additions#

Easy-access functions (scirpy.get)#

Closes issue https://github.com/scverse/scirpy/issues/184

New functions:

  • get.airr

  • get.obs_context

  • get.airr_context

Fixes#

  • Several type hints that were previously inaccurate are now updated.

  • Fix x-axis labelling in pl.clonotype_overlap raises an error if row annotations are not unique for each group.

Documentation#

The documentation has been updated to reflect the changes described above, in particular the tutorials and the page about the data structure.

Moreover, the documentation now uses a new design and moved from GitHub pages to ReadTheDocs.org. Older versions of the documentation are still accessible from github pages.

Other changes#

  • Scirpy now adopts the cookiecutter-scverse template. The structure of this repository has ben adapted accordingly. Also code was reformatted in accordance with the template defaults.

  • The minimum required Python version is now 3.9 in accordance with NEP 29

  • Increased the minium version of tqdm to 4.63 (See https://github.com/tqdm/tqdm/issues/1082)

  • pl.repertoire_overlap now always runs tl.repertoire_overlap internally and doesn’t rely on cached values.

  • The mode dendro_only in pl.repertoire_overlap has been removed.

  • Cells that have a receptor, but no CDR3 sequence have previously received a separate clonotype in tl.define_clonotypes. Now they are receiving no clonotype (i.e. np.nan) as do cells without a receptor.

  • The function tl.clonal_expansion now returns a pd.Series instead of a np.array with inplace=False

  • Removed deprecation for clonotype_imbalanced, see https://github.com/scverse/scirpy/issues/330

  • The group_abundance tool and plotting function used has_ir as a default group as we could previously rely on this column being present. With the new datastructure, this is not the case. To no break old code, the has_ir column is tempoarily added when requested. The group_abundance function will have to be rewritten enitrely in the future, see https://github.com/scverse/scirpy/issues/232

  • In pl.spectratype, the parameter groupby has been replaced by chain.

  • We now use isort to organize imports.

  • Static typing has been improved internally (using pylance). It’s not perfectly consistent yet, but we will keep working on this in the future.

  • Fix IEDB data loader after update of IEDB data formats (#401) and add tests for database import functions.

  • io.read_airr now tolerates if fields required according to the AIRR standard are missing. The respective fields will be initalized with None (#407 by @zktuong).

v0.12.2#

Fixes#

  • Fix IEDB data loader after update of IEDB data formats (backport of #401)

v0.12.1#

Fixes#

  • Bump min Python version to 3.8; CI update by @grst in https://github.com/scverse/scirpy/pull/381

  • Temporarily pin pandas < 2 in #390

Other Changes#

  • update pre-commit CI

v0.12.0#

New Features#

  • Download IEDB and process it into an AnnData object by @ausserh in https://github.com/scverse/scirpy/pull/377

Fixes#

  • Fix working with subplots (#378) by @grst in https://github.com/scverse/scirpy/pull/379

Documentation#

  • Fix typos in IR query by @Zethson in https://github.com/scverse/scirpy/pull/374

  • Fix a bunch of typos in the docs by @grst in https://github.com/scverse/scirpy/pull/375

Internal changes#

  • Fix CI by @grst in https://github.com/scverse/scirpy/pull/376

New Contributors#

  • @Zethson made their first contribution in https://github.com/scverse/scirpy/pull/374

  • @ausserh made their first contribution in https://github.com/scverse/scirpy/pull/377

Full Changelog: https://github.com/scverse/scirpy/compare/v0.11.2…v0.12.0

v0.11.2#

Fixes#

  • Excluded broken python-igraph version (#366)

v0.11.1#

Fixes#

  • Solve incompatibility with scipy v1.9.0 (#360)

Internal changes#

  • do not autodeploy docs via CI (currently broken)

  • updated patched version of scikit-learn

v0.11.0#

Additions#

  • Add data loader for BD Rhapsody single-cell immune-cell receptor data (io.read_bd_rhapsody) (#351)

Fixes#

  • Fix type conversions in from_dandelion (#349).

  • Update minimal dandelion version

Documentation#

  • Rebranding to scverse (#324, #326)

  • Add issue templates

  • Fix IMGT typos (#344 by @emjbishop)

Internal changes#

  • Bump default CI python version to 3.9

  • Use patched version of scikit-bio in CI until https://github.com/biocore/scikit-bio/pull/1813 gets merged

v0.10.1#

Fixes#

  • Fix bug in cellranger import (#310 by @ddemaeyer)

  • Fix that VDJDB download failed when cache dir was not present (#311)

v0.10.0#

Additions#

This release adds a new feature to query reference databases (#298) comprising

  • an extension of pp.ir_dist to compute distances to a reference dataset,

  • tl.ir_query, to match immune receptors to a reference database based on the distances computed with ir_dist,

  • tl.ir_query_annotate and tl.ir_query_annotate_df to annotate cells based on the result of tl.ir_query, and

  • datasets.vdjdb which conveniently downloads and processes the latest version of VDJDB.

Fixes#

  • Bump minimal dependencies for networkx and tqdm (#300)

  • Fix issue with repertoire_overlap (Fix #302 via #305)

  • Fix issue with define_clonotype_clusters (Fix #303 via #305)

  • Suppress FutureWarnings from pandas in tutorials (#307)

Internal changes#

  • Update sphinx to >= 4.1 (#306)

  • Update black version

  • Update the internal folder structure: tl, pp etc. are now real packages instead of aliases

v0.9.1#

Fixes#

  • Scirpy can now import additional columns from Cellranger 6 (#279 by @naity)

  • Fix minor issue with include_fields in AirrCell (#297)

Documentation#

  • Fix broken link in README (#296)

  • Add developer documentation (#294)

v0.9.0#

Additions#

  • Add the new “clonotype modularity” tool which ranks clonotypes by how strongly connected their gene expression neighborhood graph is. (#282).

The below example shows three clonotypes (164, 1363, 942), two of which consist of cells that are transcriptionally related.

example clonotypes clonotype modularity vs. FDR

Deprecations#

  • tl.clonotype_imbalance is now deprecated in favor of the new clonotype modularity tool.

Fixes#

  • Fix calling locus from gene name in some cases (#288)

  • Compatibility with networkx>=2.6 (#292)

Minor updates#

  • Fix some links in README (#284)

  • Fix old instances of clonotype in docs (should be clone_id) (#287)

v0.8.0#

Additions#

  • tl.alpha_diversity now supports all metrics from scikit-bio, the D50 metric and custom callback functions (#277 by @naity)

Fixes#

  • Handle input data with “productive” chains which don’t have a junction_aa sequence annotated (#281)

  • Fix issue with serialized “extra chains” not being imported correctly (#283 by @zktuong)

Minor changes#

  • The CI can now build documentation from pull-requests from forks. PR docs are not deployed to github-pages anymore, but can be downloaded as artifact from the CI run.

v0.7.1#

Fixes#

  • Ensure Compatibility with latest version of dandelion (e78701c)

  • Add links to older versions of documentation (#275)

  • Fix issue, where clonotype analysis couldn’t be continued after saving and reloading h5ad object (#274)

  • Allow “None” values to be present as cell-level attributes during merge_airr_chains (#273)

Minor changes#

  • Require anndata >= 0.7.6 in conda tests (#266)

v0.7.0#

This update features a

  • change of Scirpy’s data structure to improve interoperability with the AIRR standard

  • a complete re-write of the clonotype definition module for improved performance.

This required several backwards-incompatible changes. Please read the release notes below and the updated tutorials.

Backwards-incompatible changes#

Improve Interoperability by fully supporting the AIRR standard (#241)#

Scirpy stores receptor information in adata.obs. In this release, we updated the column names to match the AIRR Rearrangement standard. Our data model is now much more flexible, allowing to import arbitrary immune-receptor (IR)-chain related information. Use scirpy.io.upgrade_schema() to update existing AnnData objects to the latest format.

Closed issues #240, #253, #258, #255, #242, #215.

This update includes the following changes:

  • IrCell is now replaced by AirrCell which has additional functionality

  • IrChain has been removed. Use a plain dictionary instead.

  • CDR3 information is now read from the junction and junction_aa columns instead of cdr3_nt and cdr3, respectively.

  • Clonotype assignments are now per default stored in the clone_id column.

  • expr and expr_raw are now duplicate_count and consensus_count.

  • {v,d,j,c}_gene is now {v,d,j,c}_call.

  • There’s now an extra_chains column containing all IR-chains that don’t fit into our receptor model. These chains are not used by scirpy, but can be re-exported to different formats.

  • merge_with_ir is now split up into merge_with_ir (to merge IR data with transcriptomics data) and merge_airr_chains (to merge several adatas with IR information, e.g. BCR and TCR data).

  • Tutorial and documentation updates, to reflect these changes

  • Sequences are not converted to upper case on import. Scirpy tools that consume the sequences convert them to upper case on-the-fly.

  • {to,from}_ir_objs has been renamed to {to,from}_airr_cells.

Refactor CDR3 network creation (#230)#

Previously, pp.ir_neighbors constructed a cell x cell network based on clonotype similarity. This led to performance issues with highly expanded clonotypes (i.e. thousands of cells with exactly the same receptor configuration). Such cells would form dense blocks in the sparse adjacency matrix (see issue #217). Another downside was that expensive alignment-distances had to be recomputed every time the parameters of ir_neighbors was changed.

The new implementation computes distances between all unique receptor configurations, only considering one instance of highly expanded clonotypes.

Closed issues #243, #217, #191, #192, #164.

This update includes the following changes:

  • pp.ir_neighbors has been replaced by pp.ir_dist.

  • The options receptor_arms and dual_ir have been moved from pp.ir_neighbors to tl.define_clonotypes and tl.define_clonotype_clusters.

  • The default key for clonotype clusters is now cc_{distance}_{metric} instead of ct_cluster_{distance}_{metric}.

  • same_v_gene now fully respects the options dual_ir and receptor_arms

  • v-genes and receptor types were previously simply appended to clonotype ids (when same_v_gene=True). Now clonotypes with different v-genes get assigned a different numeric id.

  • Distance metric classes have been moved from ir_dist to ir_dist.metrics.

  • Distances matrices generated by ir_dist are now square and symmetric instead of triangular.

  • The default value for dual_ir is now any instead of primary_only (Closes #164).

  • The API of clonotype_network has changed.

  • Clonotype network now visualizes cells with identical receptor configurations. The number of cells with identical receptor configurations is shown as point size (and optionally, as color). Clonotype network does not support plotting multiple colors at the same time any more.

Clonotype network (previous implementation)

Clonotype network (now)

Each dot represents a cell. Cells with identical receptors form a fully connected subnetwork

Each dot represents cells with identical receptors. The dot size refers to the number of cells

image

image

Drop Support for Python 3.6#

  • Support Python 3.9, drop support for Python 3.6, following the numpy guidelines. (#229)

Fixes#

  • tl.clonal_expansion and tl.clonotype_convergence now respect cells with missing receptors and return nan for those cells. (#252)

Additions#

  • util.graph.igraph_from_sparse_matrix allows to convert a sparse connectivity or distance matrix to an igraph object.

  • ir_dist.sequence_dist now also works sequence arrays that contain duplicate entries (#192)

  • from_dandelion and to_dandelion facilitate interaction with the Dandelion package (#240)

  • write_airr allows to write scirpy’s adata.obs back to the AIRR Rearrangement format.

  • read_airr now tries to infer the locus from gene names, if no locus column is present.

  • ir.io.upgrade_schema allows to upgrade an existing scirpy anndata object to be compatible with the latest version of scirpy

  • define_clonotypes and define_clonotype_clusters now prints a logging message indicating where the results have been stored (#215)

Minor changes#

  • tqdm now uses IPython widgets to display progress bars, if available

  • the process_map from tqdm is now used to display progress bars for parallel computations instead the custom implementation used previously f307c2b

  • matplotlibs “grid lines” are now suppressed by default in all plots.

  • Docs from the master branch are now deployed to icbi-lab.github.io/scirpy/develop instead of the main documentation website. The main website only gets updated on releases.

  • Refactored the _is_na function that checks if a string evaluates to None.

  • Fixed outdated documentation of the receptor_arms parameter (#264)

v0.6.1#

Fixes#

  • Fix an issue where define_clonotype failed when the clonotype network had no edges (#236).

  • Require pandas >= 1.0 and fix a pandas incompatibility in merge_with_ir (#238).

  • Ensure consistent order of the spectratype dataframe (#238).

Minor changes#

  • Fix missing bibtex_bibfiles option in sphinx configuration

  • Work around https://github.com/takluyver/flit/issues/383.

v0.6.0#

Backwards-incompatible changes:#

  • Set more sensible defaults the the cutoff parameter in ir_neighbors. The default is now 2 for hamming and levenshtein distance metrics and 10 for the alignment distance metric.

Additions:#

  • Add Hamming-distance as additional distance metric for ir_neighbors (#216 by @ktpolanski)

Minor changes:#

  • Fix MacOS CI (#221)

  • Use mamba instead of conda in CI (#216)

v0.5.0 - Add support for BCRs and gamma-delta TCRs#

Backwards-incompatible changes:#

  • The data structure has changed. Column have been renamed from TRA_xxx and TRB_xxx to IR_VJ_xxx and IR_VDJ_xxx. Additionally a locus column has been added for each chain.

  • All occurences of tcr in the function and class names have been replaced with ir. Aliases for the old names have been created and emit a FutureWarning.

Additions:#

  • There’s now a mixed TCR/BCR example dataset (maynard2020) available (#211)

  • BCR-related amendments to the documentation (#206)

  • tl.chain_qc which supersedes chain_pairing. It additionally provides information about the receptor type.

  • io.read_tracer now supports gamma-delta T-cells (#207)

  • io.to_ir_objs allows to convert adata to a list of IrCells (#210)

  • io.read_bracer allows to read-in BraCeR BCR data. (#208)

  • The pp.merge_with_ir function now can handle the case when both the left and the right AnnData object contain immune receptor information. This is useful when integrating both TCR and BCR data into the same dataset. (#210)

Fixes:#

  • Fix a bug in vdj_usage which has been triggered by the new data structure (#203)

Minor changes:#

  • Removed the tqdm monkey patch, as the issue has been resolved upstream (#200)

  • Add AIRR badge, as scirpy is now certified to comply with the AIRR software standard v1. (#202)

  • Require pycairo >1.20 which provides a windows wheel, eliminating the CI problems.

v0.4.2#

  • Include tests into main package (#189)

v0.4.1#

  • Fix pythonpublish CI action

  • Update black version (and code style, accordingly)

  • Changes for AIRR-complicance:

    • Add support level to README

    • Add Biocontainer instructions to README

    • Add a minimal test suite to be ran on conda CI

v0.4#

  • Adapt tcr_dist to support second array of sequences (#166). This enables comparing CDR3 sequences against a list of reference sequences.

  • Add tl.clonotype_convergence which helps to find evidence of convergent evolution (#168)

  • Optimize parallel sequence distance calculation (#171). There is now less communication overhead with the worker processes.

  • Fixed an error when runing pp.tcr_neighbors (#177)

  • Improve packaging. Use setuptools_scm instead of get_version. Remove redundant metadata. (#180). More tests for conda (#180).

v0.3#

  • More extensive CI tests (now also testing on Windows, MacOS and testing the conda recipe) (#136, #138)

  • Add example images to API documentation (#140)

  • Refactor IO to expose TcrCell and TcrChain (#139)

  • Create data loading tutorial (#139)

  • Add a progressbar to TCR neighbors (#143)

  • Move clonotype_network_igraph to tools (#144)

  • Add read_airr to support the AIRR rearrangement format (#147)

  • Add option to take v-gene into account during clonotype definition (#148)

  • Store colors in AnnData to ensure consistent coloring across plots (#151)

  • Divide define_clontoypes into define_clonotypes and define_clonotype_clusters (#152). Now, the user has to specify explicitly sequence and metric for both tl.tcr_neighbors, tl.define_clonotype_clusters and tl.clonotype_network. This makes it more straightforward to have multiple, different versions of the clonotype network at the same time. The default parameters changed to sequence="nt" and `metric=”identity” to comply with the traditional definition of clonotypes. The changes are also reflected in the glossary and the tutorial.

  • Update the workflow figure (#154)

  • Fix a bug that caused labels in the repertoire_overlap heatmap to be mixed up. (#157)

  • Add a label to the heatmap annotation in repertoire_overlap (#158).

v0.2#

  • Documentation overhaul. A lot of docstrings got corrected and improved and the formatting of the documentation now matches scanpy’s.

  • Experimental function to assess bias in clonotype abundance between conditions (#92)

  • Scirpy now has a logo (#123)

  • Update default parameters for clonotype_network:

    • Edges are now only automatically displayed if plotting < 1000 nodes

    • If plotting variables with many categories, the legend is hidden.

  • Update default parameters for alignment-based tcr_neighbors

    • The gap extend penalty now equals the gap open penalty (11).

v0.1.2#

  • Make 10x csv and json import consistent (#109)

  • Fix version requirements (#112)

  • Fix compatibility issues with pandas > 1 (#112)

  • Updates to tutorial and README

v0.1.1#

  • Update documentation about T-cell receptor model (#4, #10)

  • Update README

  • Fix curve plots (#31)

  • Host datasets on GitHub (#104)

v0.1#

Initial release for pre-print