Use gtfparse 3 with explicit pandas 3 coverage

.github/workflows/tests.yml

@@ -38,7 +38,15 @@ jobs:
     strategy:
       fail-fast: true
       matrix:
-        python-version: ["3.10", "3.11", "3.12"]
+        include:
+          - python-version: "3.10"
+            pandas: "pandas>=2.2.2,<3"
+          - python-version: "3.11"
+            pandas: "pandas>=2.2.2,<3"
+          - python-version: "3.11"
+            pandas: "pandas>=3,<4"
+          - python-version: "3.12"
+            pandas: "pandas>=3,<4"
     steps:
       - uses: actions/checkout@v7
       - name: Set up Python ${{ matrix.python-version }}
@@ -74,10 +82,14 @@ jobs:
           echo `which netMHCpan` && netMHCpan -h
           echo `which netMHCcons` && netMHCcons -h
       - name: Install dependencies
+        env:
+          PANDAS_REQUIREMENT: ${{ matrix.pandas }}
         run: |
           python -m pip install --upgrade pip
           python -m pip install pytest pytest-cov ruff
-          bash scripts/pip_install.sh -e .
+          bash scripts/pip_install.sh -e . "$PANDAS_REQUIREMENT"
+          python -m pip check
+          python -m pip show pandas pyensembl gtfparse
       - name: Resolve the installed default Ensembl release
         id: ensembl-release
         run: |
@@ -166,7 +178,7 @@ jobs:
       - name: Save coverage for the upload job
         uses: actions/upload-artifact@v7
         with:
-          name: coverage-python-${{ matrix.python-version }}
+          name: coverage-python-${{ matrix.python-version }}-${{ strategy.job-index }}
           path: .coverage
           include-hidden-files: true
           if-no-files-found: error

CHANGELOG.md

@@ -7,6 +7,14 @@ between published tags; older pre-5.0 notes are retained below.
 
 For current interfaces, see the [consumer guide](docs/consumer-guide.md).
 
+## 5.94.2
+
+- Require gtfparse 3.0.2+ and PyEnsembl 2.24.1+, allowing the current pandas 3
+  annotation stack alongside Varcode (#484). Retain pandas 2.2.2+ for Python
+  3.10 and explicitly test both pandas 2 and 3 on Python 3.11.
+- Select pandas output in both GTF expression readers and verify their
+  transcript FPKM values against the same StringTie fixture.
+
 ## 5.94.1
 
 - Apply explicit Cufflinks HIDATA expression replacements under pandas

requirements.txt

@@ -1,10 +1,10 @@
 numpy>=2.0.0,<3.0.0
-pandas>=2.0.0
+pandas>=2.2.2,<4.0.0
 mhctools>=3.45.1
 varcode>=4.18.0
-gtfparse>=2.8.1,<3.0.0
+gtfparse>=3.0.2,<4.0.0
 mhcgnomes>=3.4.0
-pyensembl>=2.3.0,<3.0.0
+pyensembl>=2.24.1,<3.0.0
 # PEP 440 versions and validated package names for release preflight
 packaging>=23.2
 argcomplete>=1.10

tests/test_load_stringtie_gtf_fpkm.py

@@ -1,11 +1,15 @@
-from topiary.rna import load_transcript_fpkm_dict_from_gtf
+"""Both GTF readers preserve known abundances with the pandas parser backend."""
+
+import pytest
 
 from .common import eq_
 from .data import data_path
+from .test_twin_conformance import GTF_FPKM_TWINS
 
 
-def test_load_stringtie_gtf_transcripts():
-    transcript_fpkms = load_transcript_fpkm_dict_from_gtf(
+@pytest.mark.parametrize("name,reader", GTF_FPKM_TWINS, ids=[name for name, _ in GTF_FPKM_TWINS])
+def test_load_stringtie_gtf_transcripts(name, reader):
+    transcript_fpkms = reader(
         data_path("B16-StringTie-chr1-subset.gtf")
     )
     transcript_ids = set(transcript_fpkms.keys())
@@ -18,27 +22,3 @@ def test_load_stringtie_gtf_transcripts():
     eq_(expected_transcript_ids, transcript_ids)
     for transcript_id, fpkm in expected_fpkms_dict.items():
         eq_(fpkm, transcript_fpkms[transcript_id])
-
-
-def test_load_expression_reads_a_gtf_as_pandas():
-    """The other GTF door, which had no test and was broken.
-
-    ``topiary.rna.gtf`` and ``expression_loader._load_gtf`` both read
-    GTFs, and only the first was covered. gtfparse 2.x returns polars by
-    default; ``_load_gtf`` treats the result as pandas, so every GTF
-    read through ``load_expression`` (and through the CLI's
-    ``--transcript-expression``) raised "expected 17 values when
-    selecting columns by boolean mask" from polars, while this file's
-    other test kept passing.
-    """
-    import pandas as pd
-
-    from topiary.rna.expression_loader import load_expression
-
-    df = load_expression(data_path("B16-StringTie-chr1-subset.gtf"))
-
-    assert isinstance(df, pd.DataFrame)
-    assert not df.empty
-    # Transcript rows only, carrying the abundance column.
-    assert "reference_id" in df.columns
-    assert "TPM" in df.columns or "FPKM" in df.columns

tests/test_twin_conformance.py

@@ -41,6 +41,8 @@
     attach_rna_evidence,
 )
 from topiary.rna import load_cufflinks_dataframe, load_cufflinks_dict, load_cufflinks_fpkm_dict
+from topiary.rna import load_transcript_fpkm_dict_from_gtf
+from topiary.rna.expression_loader import load_expression
 from topiary import (
     APPROXIMATED, MEASURED, CachedPredictor, ProteinFragment, TopiaryPredictor, from_predictions, fragments_from_variants,
     read_fragments, read_pvacseq, write_fragments, unique_fragments,
@@ -125,6 +127,14 @@ def calls(self, arg, value):
 )
 
 
+# Both public StringTie readers must retain the same transcript FPKM values.
+GTF_FPKM_TWINS = (
+    ("values", load_transcript_fpkm_dict_from_gtf),
+    ("dataframe", lambda path: load_expression(
+        path, val_cols="FPKM").set_index("reference_id")["FPKM"].to_dict()),
+)
+
+
 # Drive both input forms through the same hypothesis validation battery.
 ISOVAR_HYPOTHESIS_INPUT_TWINS = (
     ("mapping", lambda export, path: read_isovar_hypotheses(export, tag="hypotheses")),

topiary/__init__.py

@@ -208,7 +208,7 @@
     encode_amino_acids,
 )
 
-__version__ = "5.94.1"
+__version__ = "5.94.2"
 
 __all__ = [
     "normalize_isovar_rna_support",

topiary/rna/expression_loader.py

@@ -139,11 +139,8 @@ def _load_gtf(filepath, id_col=None, val_cols=None):
     """Load expression data from a StringTie GTF file."""
     import gtfparse
 
-    # result_type is required, not cosmetic: gtfparse 2.x defaults to
-    # returning polars, and everything below treats df as pandas. Without
-    # it, reading any GTF raises "expected 17 values when selecting
-    # columns by boolean mask" from polars on the next line. The keyword
-    # exists only from gtfparse 2.7, which requirements.txt now floors.
+    # Keep the backend explicit: filtering and numeric conversion below
+    # use pandas, which is also gtfparse 3's default result type.
     df = gtfparse.read_gtf(filepath, result_type="pandas")
     # Filter to transcript features only
     if "feature" in df.columns:

topiary/rna/gtf.py

@@ -41,7 +41,8 @@ def load_transcript_fpkm_dict_from_gtf(
     quantification of abundance. Returns a dictionary mapping Ensembl
     IDs of transcripts to FPKM values.
     """
-    df = gtfparse.read_gtf(gtf_path, column_converters={fpkm_column_name: float})
+    df = gtfparse.read_gtf(
+        gtf_path, column_converters={fpkm_column_name: float}, result_type="pandas")
     transcript_ids = _get_gtf_column(transcript_id_column_name, gtf_path, df)
     fpkm_values = _get_gtf_column(fpkm_column_name, gtf_path, df)
     features = _get_gtf_column(feature_column_name, gtf_path, df)