[alpae]

Part of #43: the pyham upgrade that should land before the algorithm fixes (#36, #37, #38).

Dependencies

• pyham>=2.0.0rc1,<3 (to be raised to >=2.0.0 once pyham 2.0.0 is released) and ete4>=4.3,<5 instead of ete3
• version 0.2.0
• ete4 4.4.0 is only distributed as source (Cython), so installing it needs a C compiler. The Docker builder stage now installs gcc. The oma extra still pulls ete3 via pyoma; both packages can be installed side by side.

ete4 props

ete4 stores data as props of the tree nodes and does not support attribute access (node.genome) or setting arbitrary attributes (node.n_internal_nodes = ...). Consistently throughout edgehog:

• results are stored with tree_node.add_prop(name, value) (was add_feature)
• values edgehog always sets are read with tree_node.props[name]
• the genome is read with tree_node.props.get('genome') (None replaces the former AttributeError)
• is_leaf is a property
• missing branch lengths are None in ete4 (1.0 in ete3); characterize_species_tree.branch_length counts them as 1.0, so the RED scores do not change

utils/extract_age_timetree.py is ported to ete4 as well.

Optional --species_tree

pyham 2 can read the species tree from a <taxonomy> section of the orthoXML (e.g. orthoXML exported from the OMA browser). --species_tree is therefore optional; if both are given, the species tree is used (pyham's behavior). Without either, edgehog exits with pyham's error message.

HOG ids

No change needed in edgehog. HOG ids are only used for output labels (duplicates are already made unique with _c<n>) and by the HDF5 writer, which strips the _<taxid> part to look up the OMA rows. pyham 2 gives HOGs that it synthesizes for paralogous branches their own LOFT ids, which makes the labels of in-paralogs distinguishable and the taxid lookup in get_taxid_from_hog_names more reliable.

Comparison with pyham 1

I ran test_data (GFF input, with --date_edges --orient_edges --phylostratify) and the OMA HDF5 test dataset (--hdf5, TSV and HDF5 output) with pyham 1.2.0 (current main) and with this branch, and compared all synteny graphs at every level. Each graph node was identified by its set of descendant extant proteins, so the comparison does not depend on HOG ids or node order.

• HDF5 test dataset: all graphs identical (nodes, edges, weights, orientation, dating, RED scores); Synteny.h5 identical.
• test_data: all edges and edge attributes identical. pyham 1 created 12 spurious ancestral genes without edges for families with duplications, e.g. it extended HOG:0001787_11, which the orthoXML roots at Hominidae, up to Simiiformes, and put an extra HOG at Cercopithecinae above the Simiiformes root of HOG:0001733_1. pyham 2 follows the orthoXML, so these singletons are no longer in the output.
• The TSV files still differ textually: some labels changed (see HOG ids), edges are sometimes written as (gene2, gene1), and weight is written as 4 instead of 4.0 when a graph has no singletons.

With the <taxonomy> of the OMA orthoXML instead of simplified_speciestree.nwk, all graphs are identical, except that one level is named Saccharomycetales instead of Saccharomycotina.

Tests

• tests converted to ete4 props
• test_synthetic_pipeline.py now also runs with the species tree as <taxonomy> in the orthoXML and no tree file, with the same expected ancestral orders
• new test_species_tree.py: RED scores and levels, missing branch lengths counting as 1.0, the optional --species_tree, the error without any taxonomy, and the newick taking precedence over the <taxonomy>
• the 7 xfail tests still fail for the same reasons
• full suite passes on Python 3.12 (macOS) and 3.9 (Linux container, ete4 built from source); docker build and a run of test_data in the image work