Add a changelog for 0.2.0

CHANGELOG.md

@@ -0,0 +1,94 @@
+# Changelog
+
+## 0.2.0 (unreleased)
+
+This release fixes several deviations from the algorithm described in the
+Supplementary Information of the edgeHOG paper
+([doi:10.1038/s41559-025-02818-0](https://doi.org/10.1038/s41559-025-02818-0))
+and bugs in reading the input (tracked in
+[#43](https://github.com/DessimozLab/edgehog/issues/43)). **The inferred
+ancestral synteny graphs change for existing datasets**, mostly because of
+the gene order of unsorted GFF files.
+
+### Changes in results
+- Extant genes are ordered by contig and genomic coordinates before adjacent
+  genes are linked. Before, the order of the GFF file was used, which links
+  wrong genes if the GFF is not sorted. On `test_data`, about 27% of the
+  linearized ancestral edges change.
+  ([#40](https://github.com/DessimozLab/edgehog/issues/40),
+  [#46](https://github.com/DessimozLab/edgehog/pull/46))
+- The orientation of an extant edge uses the strand of the previous linked
+  gene, not of a gene skipped because it is missing from the orthoXML.
+  ([#40](https://github.com/DessimozLab/edgehog/issues/40),
+  [#46](https://github.com/DessimozLab/edgehog/pull/46))
+- Bottom-up phase: if the edge reconnecting the neighbors of a gene without
+  parent HOG already exists, the minimal weight of the two removed edges is
+  added to it, as described in the paper. Before, it was left unchanged, and
+  the two removed edges could be pruned in the top-down phase. For gaps of
+  three or more genes, all real edges are now mapped to their reconnected
+  edge, independent of the order in which the genes are removed.
+  ([#38](https://github.com/DessimozLab/edgehog/issues/38),
+  [#51](https://github.com/DessimozLab/edgehog/pull/51))
+- Top-down phase with `--orient_edges`: orientation contexts not supported
+  by two children or by the parent are reset to 0, as described in the paper;
+  if none is supported, they stay unchanged. Before, the bottom-up contexts
+  were kept. `nb_internal_nodes_from_ancestor_with_updated_weight` no longer
+  depends on `--orient_edges`.
+  ([#36](https://github.com/DessimozLab/edgehog/issues/36),
+  [#49](https://github.com/DessimozLab/edgehog/pull/49))
+- Linearization: removed edges are reintroduced by descending weight, then by
+  descending weight of the resulting connected component (the second
+  criterion was inverted), and no longer when they would close a cycle.
+  Ties are broken in a deterministic order.
+  ([#37](https://github.com/DessimozLab/edgehog/issues/37),
+  [#50](https://github.com/DessimozLab/edgehog/pull/50))
+- With pyham 2, ancestral genes follow the HOGs of the orthoXML. pyham 1
+  added spurious ancestral genes without edges for some families with
+  duplications. Some HOG labels in the output change, as pyham 2 gives HOGs
+  of paralogous branches their own ids.
+  ([#48](https://github.com/DessimozLab/edgehog/pull/48))
+
+### Added
+- `--species_tree` is optional if the orthoXML contains a `<taxonomy>`
+  section (e.g. orthoXML exported from the OMA browser). If both are given,
+  the species tree is used.
+  ([#48](https://github.com/DessimozLab/edgehog/pull/48))
+- Support for Python 3.13 and 3.14. The `oma` extra (needed for `--hdf5`)
+  is only available for Python < 3.13, because pyoma requires numpy < 2.
+- CDS-only GFF files without `protein_id` (e.g. Prodigal output) and GFF3
+  files with a `##FASTA` section can be read. For Prodigal IDs, the protein
+  ID is `<contig>_<orf number>`, as in the headers of Prodigal's protein
+  fasta. ([#41](https://github.com/DessimozLab/edgehog/issues/41),
+  [#45](https://github.com/DessimozLab/edgehog/pull/45))
+
+### Changed
+- Depends on pyham 2 and ete4 instead of pyham 1 and ete3. ete4 is only
+  distributed as source, so installing edgehog needs a C compiler.
+  ([#48](https://github.com/DessimozLab/edgehog/pull/48))
+- GFF lookup: each GFF file is attributed to the longest genome name it
+  starts with, and `<name>.gff` / `<name>.gff3` is preferred. Only files
+  ending in `.gff` or `.gff3`, optionally compressed (`.gz`, `.bz2`, `.xz`,
+  `.zip`), are considered. If several files remain for a genome, edgehog exits
+  with an error listing them; before, it used the first one in directory
+  order. ([#42](https://github.com/DessimozLab/edgehog/issues/42),
+  [#44](https://github.com/DessimozLab/edgehog/pull/44))
+- For CDS lines with an `ID` but no `protein_id`, the ID is used as protein
+  ID (unless it is in Prodigal's format). Before, it was truncated to
+  `<contig>_<second part of the ID>`, which failed for IDs without `_`.
+  ([#45](https://github.com/DessimozLab/edgehog/pull/45))
+- In the output TSV files, the two genes of an edge can be written in either
+  order, and `weight` is written as an integer when a graph has no singletons.
+
+### Fixed
+- A missing GFF file exits with an error message instead of raising
+  `IndexError`. ([#44](https://github.com/DessimozLab/edgehog/pull/44))
+- A genome whose name is a prefix of another genome's name (`HUMAN` and
+  `HUMAN2`) could read the other genome's GFF.
+  ([#44](https://github.com/DessimozLab/edgehog/pull/44))
+
+### Development
+- Build and dependency management switched from poetry to uv.
+  ([#34](https://github.com/DessimozLab/edgehog/pull/34))
+- Unit test suite for GFF parsing and every inference phase, including the
+  example of Fig. 2 of the paper.
+  ([#35](https://github.com/DessimozLab/edgehog/pull/35))

README.md

@@ -32,7 +32,9 @@ The package development version is tested on *Linux* operating systems. The deve
 
 The package itself should be compatible with Windows, Mac and Linux operating systems.
 
-Edgehog is written in Python, so a working python installation is needed before installing edgehog. The version of python version must be >=3.9 and <3.13
+Edgehog is written in Python, so a working python installation is needed before installing edgehog. The version of python must be >=3.9 and <3.15 (<3.13 for the `oma` extra, see below).
+
+Installing edgehog also needs a C compiler (e.g. `gcc`), because its dependency `ete4` is only distributed as source code.
 
 #### Installing Python on Ubuntu 22.04
 
@@ -48,7 +50,7 @@ pip install edgehog
 ```
 
 To enable hdf5 support and direct reading of genome data from OMA's HDF5 database, you need to enable 
-the `oma` extra during installation:
+the `oma` extra during installation (only available for python <3.13, because `pyoma` requires numpy<2):
 
 ```bash
 pip install edgehog[oma]

pyproject.toml

@@ -34,6 +34,7 @@ oma = [
 
 [project.urls]
 Repository = "https://github.com/DessimozLab/edgehog"
+Changelog = "https://github.com/DessimozLab/edgehog/blob/main/CHANGELOG.md"
 
 [project.scripts]
 edgehog = 'edgehog.edgehog:main'