napistu 0.2.5.dev6__py3-none-any.whl → 0.3.1__py3-none-any.whl

napistu/ingestion/bigg.py

@@ -1,21 +1,16 @@
 from __future__ import annotations
 
-import datetime
 import logging
 import os
 from typing import Iterable
 
-import pandas as pd
 from napistu import indices
 from napistu import sbml_dfs_core
 from napistu import utils
 from napistu.consensus import construct_sbml_dfs_dict
-from napistu.ingestion import sbml
-from napistu.ingestion.constants import BIGG_MODEL_FIELD_SPECIES
-from napistu.ingestion.constants import BIGG_MODEL_FIELD_URL
+from napistu.ontologies.renaming import rename_species_ontologies
 from napistu.ingestion.constants import BIGG_MODEL_KEYS
 from napistu.ingestion.constants import BIGG_MODEL_URLS
-from napistu.ingestion.constants import BIGG_RECON3D_FIELD_ANNOTATION
 from napistu.ingestion.constants import SPECIES_FULL_NAME_HUMAN
 from napistu.ingestion.constants import SPECIES_FULL_NAME_MOUSE
 from napistu.ingestion.constants import SPECIES_FULL_NAME_YEAST
@@ -40,33 +35,17 @@ def bigg_sbml_download(bg_pathway_root: str, overwrite: bool = False) -> None:
     """
     utils.initialize_dir(bg_pathway_root, overwrite)
 
-    bigg_models = {
-        BIGG_MODEL_KEYS[SPECIES_FULL_NAME_HUMAN]: {
-            BIGG_MODEL_FIELD_URL: BIGG_MODEL_URLS[SPECIES_FULL_NAME_HUMAN],
-            BIGG_MODEL_FIELD_SPECIES: SPECIES_FULL_NAME_HUMAN,
+    bigg_models_df = indices.create_pathway_index_df(
+        model_keys=BIGG_MODEL_KEYS,
+        model_urls=BIGG_MODEL_URLS,
+        model_species={
+            SPECIES_FULL_NAME_HUMAN: SPECIES_FULL_NAME_HUMAN,
+            SPECIES_FULL_NAME_MOUSE: SPECIES_FULL_NAME_MOUSE,
+            SPECIES_FULL_NAME_YEAST: SPECIES_FULL_NAME_YEAST,
         },
-        BIGG_MODEL_KEYS[SPECIES_FULL_NAME_MOUSE]: {
-            BIGG_MODEL_FIELD_URL: BIGG_MODEL_URLS[SPECIES_FULL_NAME_MOUSE],
-            BIGG_MODEL_FIELD_SPECIES: SPECIES_FULL_NAME_MOUSE,
-        },
-        BIGG_MODEL_KEYS[SPECIES_FULL_NAME_YEAST]: {
-            BIGG_MODEL_FIELD_URL: BIGG_MODEL_URLS[SPECIES_FULL_NAME_YEAST],
-            BIGG_MODEL_FIELD_SPECIES: SPECIES_FULL_NAME_YEAST,
-        },
-    }
-    bigg_models_df = pd.DataFrame(bigg_models).T
-    bigg_models_df["sbml_path"] = [
-        os.path.join(bg_pathway_root, k) + ".sbml"
-        for k in bigg_models_df.index.tolist()
-    ]
-    bigg_models_df["file"] = [os.path.basename(x) for x in bigg_models_df["sbml_path"]]
-
-    # add other attributes which will be used in the pw_index
-    bigg_models_df["date"] = datetime.date.today().strftime("%Y%m%d")
-    bigg_models_df.index = bigg_models_df.index.rename("pathway_id")
-    bigg_models_df = bigg_models_df.reset_index()
-    bigg_models_df["name"] = bigg_models_df["pathway_id"]
-    bigg_models_df = bigg_models_df.assign(source="BiGG")
+        base_path=bg_pathway_root,
+        source_name="BiGG",
+    )
 
     with open_fs(bg_pathway_root, create=True) as bg_fs:
         for _, row in bigg_models_df.iterrows():
@@ -84,41 +63,46 @@ def bigg_sbml_download(bg_pathway_root: str, overwrite: bool = False) -> None:
     return None
 
 
-def annotate_recon(raw_model_path: str, annotated_model_path: str) -> None:
-    """Annotate Recon3D
-    Add compartment annotations to Recon3D so it can be merged with other pathways
-    """
-    logger.warning(
-        "add_sbml_annotations is deprecated and maybe removed in a future version of rcpr; "
-        "we are now adding these annotation during ingestion by sbml.sbml_df_from_sbml() rather "
-        "than directly appending them to the raw .sbml"
-    )
-    recon_3d_annotations = pd.DataFrame(BIGG_RECON3D_FIELD_ANNOTATION)
-    sbml_model = sbml.SBML(raw_model_path)
-    sbml.add_sbml_annotations(
-        sbml_model, recon_3d_annotations, save_path=annotated_model_path
-    )
-
-    return None
-
-
 def construct_bigg_consensus(
     pw_index_inp: str | indices.PWIndex,
     species: str | Iterable[str] | None = None,
     outdir: str | None = None,
 ) -> sbml_dfs_core.SBML_dfs:
-    """Constructs a BiGG SBML DFs Pathway Representation
-
-    Attention: curently this does work only for a singly model. Integraiton of multiple
-    models is not supported yet in BiGG.
-
-    Args:
-        pw_index_inp (str | indices.PWIndex): PWIndex or uri pointing to PWIndex
-        species (str | Iterable[str] | None): one or more species to filter by. Default: no filtering
-        outdir (str | None, optional): output directory used to cache results. Defaults to None.
-
-    Returns:
-        sbml_dfs_core.SBML_dfs: A consensus SBML
+    """Construct a BiGG SBML DFs pathway representation.
+
+    Parameters
+    ----------
+    pw_index_inp : str or indices.PWIndex
+        PWIndex object or URI pointing to PWIndex
+    species : str or Iterable[str] or None, optional
+        One or more species to filter by, by default None (no filtering)
+    outdir : str or None, optional
+        Output directory used to cache results, by default None
+
+    Returns
+    -------
+    sbml_dfs_core.SBML_dfs
+        A consensus SBML representation
+
+    Notes
+    -----
+    Currently this only works for a single model. Integration of multiple
+    models is not yet supported in BiGG.
+
+    The function:
+    1. Loads/validates the pathway index
+    2. Constructs SBML DFs dictionary
+    3. Processes the single model:
+        - Infers compartmentalization for species without location
+        - Names compartmentalized species
+        - Validates the final model
+
+    Raises
+    ------
+    ValueError
+        If pw_index_inp is neither a PWIndex nor a string
+    NotImplementedError
+        If attempting to merge multiple models
     """
     if isinstance(pw_index_inp, str):
         pw_index = indices.adapt_pw_index(pw_index_inp, species=species, outdir=outdir)
@@ -142,5 +126,6 @@ def construct_bigg_consensus(
     # fix missing compartimentalization
     model = sbml_dfs_core.infer_uncompartmentalized_species_location(model)
     model = sbml_dfs_core.name_compartmentalized_species(model)
+    rename_species_ontologies(model)
     model.validate()
     return model

napistu/ingestion/constants.py

@@ -3,12 +3,30 @@ from __future__ import annotations
 
 from types import SimpleNamespace
 
+
 SPECIES_FULL_NAME_HUMAN = "Homo sapiens"
 SPECIES_FULL_NAME_MOUSE = "Mus musculus"
 SPECIES_FULL_NAME_YEAST = "Saccharomyces cerevisiae"
 SPECIES_FULL_NAME_RAT = "Rattus norvegicus"
 SPECIES_FULL_NAME_WORM = "Caenorhabditis elegans"
 
+PROTEINATLAS_SUBCELL_LOC_URL = (
+    "https://www.proteinatlas.org/download/tsv/subcellular_location.tsv.zip"
+)
+
+PROTEINATLAS_DEFS = SimpleNamespace(
+    GO_ID="GO id",
+    GENE="Gene",
+)
+
+
+# GTEx
+GTEX_RNASEQ_EXPRESSION_URL = "https://storage.googleapis.com/adult-gtex/bulk-gex/v8/rna-seq/GTEx_Analysis_2017-06-05_v8_RNASeQCv1.1.9_gene_median_tpm.gct.gz"
+
+GTEX_DEFS = SimpleNamespace(
+    NAME="Name",
+    DESCRIPTION="Description",
+)
 
 # BIGG
 BIGG_MODEL_URLS = {
@@ -29,134 +47,6 @@ BIGG_RECON3D_FIELD_ID = "id"
 BIGG_RECON3D_FIELD_TYPE = "type"
 BIGG_RECON3D_FIELD_URI = "uri"
 
-BIGG_RECON3D_ID_C = "c"
-BIGG_RECON3D_ID_L = "l"
-BIGG_RECON3D_ID_E = "e"
-BIGG_RECON3D_ID_M = "m"
-BIGG_RECON3D_ID_R = "r"
-BIGG_RECON3D_ID_X = "x"
-BIGG_RECON3D_ID_N = "n"
-BIGG_RECON3D_ID_I = "i"
-
-BIGG_RECON3D_TYPE_COMPARTMENT = "compartment"
-
-BIGG_RECON3D_FIELD_ANNOTATION = [
-    {
-        # cytosol
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_C,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005829",
-    },
-    {
-        # cytoplasm
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_C,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005737",
-    },
-    {
-        # plasma membrane
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_C,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005886",
-    },
-    {
-        # lysosome lumen
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_L,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0043202",
-    },
-    {
-        # lysosomal membrane
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_L,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005765",
-    },
-    {
-        # mitochondrial intermembrane space
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_M,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005758",
-    },
-    {
-        # mitochondrial outer membrane
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_M,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005741",
-    },
-    {
-        # ER membrane
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_R,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005789",
-    },
-    {
-        # ER lumen
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_R,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005788",
-    },
-    {
-        # extracellular region
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_E,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005576",
-    },
-    {
-        # peroxosomal membrane
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_X,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005778",
-    },
-    {
-        # peroxosomal matrix
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_X,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005782",
-    },
-    {
-        # nucleolus
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_N,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005730",
-    },
-    {
-        # nuclear envelope
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_N,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005635",
-    },
-    {
-        # nucleoplasm
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_N,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005654",
-    },
-    {
-        # golgi membrane
-        BIGG_RECON3D_FIELD_ID: "g",
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0000139",
-    },
-    {
-        # golgi lumen
-        BIGG_RECON3D_FIELD_ID: "g",
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005796",
-    },
-    {
-        # mitochondrial matrix
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_I,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005759",
-    },
-    {
-        # mitochondrial inner membrane
-        BIGG_RECON3D_FIELD_ID: BIGG_RECON3D_ID_I,
-        BIGG_RECON3D_FIELD_TYPE: BIGG_RECON3D_TYPE_COMPARTMENT,
-        BIGG_RECON3D_FIELD_URI: "https://www.ebi.ac.uk/QuickGO/term/GO:0005743",
-    },
-]
-
 # IDENTIFIERS ETL
 IDENTIFIERS_ETL_YEAST_URL = "https://www.uniprot.org/docs/yeast.txt"
 IDENTIFIERS_ETL_SBO_URL = (
@@ -239,11 +129,6 @@ SBML_COMPARTMENTALIZED_SPECIES_DICT_SOURCE = "sc_Source"
 
 SBML_REACTION_ATTR_GET_GENE_PRODUCT = "getGeneProduct"
 
-SBML_ANNOTATION_METHOD_GET_SPECIES = "getSpecies"
-SBML_ANNOTATION_METHOD_GET_COMPARTMENT = "getCompartment"
-SBML_ANNOTATION_METHOD_GET_REACTION = "getReaction"
-
-
 # STRING
 STRING_URL_EXPRESSIONS = {
     "interactions": "https://stringdb-static.org/download/protein.links.full.v{version}/{taxid}.protein.links.full.v{version}.txt.gz",

napistu/ingestion/gtex.py

@@ -0,0 +1,113 @@
+from __future__ import annotations
+
+import logging
+import pandas as pd
+from fs import open_fs
+from napistu import utils
+
+from napistu.constants import ONTOLOGIES
+from napistu.ingestion.constants import GTEX_DEFS, GTEX_RNASEQ_EXPRESSION_URL
+
+logger = logging.getLogger(__name__)
+
+
+def download_gtex_rnaseq(
+    target_uri: str, url: str = GTEX_RNASEQ_EXPRESSION_URL
+) -> None:
+    """Download GTEx RNA-seq expression data.
+
+    Parameters
+    ----------
+    target_uri : str
+        The URI where the GTEx data should be saved
+    url : str, optional
+        URL to download the GTEx RNA-seq expression data from.
+        Defaults to GTEX_RNASEQ_EXPRESSION_URL.
+
+    Returns
+    -------
+    None
+
+    Notes
+    -----
+    Downloads GTEx RNA-seq expression data (median TPM per gene per tissue) from the
+    specified URL and saves it to the target URI. By default, downloads from GTEx
+    Analysis V8 data (dbGaP Accession phs000424.v8.p2).
+    """
+    logger.info("Start downloading gtex %s to %s", url, target_uri)
+    utils.download_wget(url, target_uri)
+
+
+def load_and_clean_gtex_data(gtex_data_path: str) -> pd.DataFrame:
+    """Load and format GTEx tissue specific expression data.
+
+    This function loads tissue-specific expression data from GTEx (median value per gene per tissue).
+
+    Parameters
+    ----------
+    gtex_data_path : str
+        Path to GTEx tissue specific expression data (medians)
+
+    Returns
+    -------
+    pd.DataFrame
+        DataFrame containing all the information from the GTEx file with standardized column names:
+        - ensembl_gene_id: Ensembl gene ID without version number
+        - ensembl_geneTranscript_id: Original GTEx hybrid gene/transcript ID
+        - Description: Gene description/symbol
+        - Multiple tissue columns with median TPM values
+
+    Notes
+    -----
+    The function:
+    1. Skips the first 2 lines of the GTEx file (header info)
+    2. Creates clean Ensembl gene IDs by removing version numbers
+    3. Renames columns for clarity
+    4. Reorders columns to put ID and description columns first
+
+    Raises
+    ------
+    FileNotFoundError
+        If the input file does not exist
+    """
+    # Check file exists
+    base_path, file_name = utils.get_source_base_and_path(gtex_data_path)
+
+    logger.info("Loading GTEx tissue specific expression data")
+
+    # Read the TSV file using pandas, skipping first 2 lines
+    with open_fs(base_path) as base_fs:
+        with base_fs.open(file_name, "rb") as f:
+            gtex_expression_data = pd.read_csv(
+                f, sep="\t", skiprows=2, dtype=str, na_values=[""], keep_default_na=True
+            )
+
+    # Create ensembl_gene_id by removing version numbers from Name column
+    gtex_expression_data[ONTOLOGIES.ENSEMBL_GENE] = gtex_expression_data[
+        GTEX_DEFS.NAME
+    ].str.replace(r"\.[0-9]+$", "", regex=True)
+
+    # Rename Name column to be more informative
+    gtex_expression_data = gtex_expression_data.rename(
+        columns={
+            GTEX_DEFS.NAME: ONTOLOGIES.ENSEMBL_GENE_VERSION,
+            GTEX_DEFS.DESCRIPTION: ONTOLOGIES.SYMBOL,
+        }
+    )
+
+    # Reorder columns to put ID and description columns first
+    first_cols = [
+        ONTOLOGIES.ENSEMBL_GENE,
+        ONTOLOGIES.ENSEMBL_GENE_VERSION,
+        ONTOLOGIES.SYMBOL,
+    ]
+    other_cols = [col for col in gtex_expression_data.columns if col not in first_cols]
+    gtex_expression_data = gtex_expression_data[first_cols + other_cols]
+
+    # Convert tissue columns to numeric
+    numeric_cols = [col for col in other_cols if col not in first_cols]
+    gtex_expression_data[numeric_cols] = gtex_expression_data[numeric_cols].apply(
+        pd.to_numeric, errors="coerce"
+    )
+
+    return gtex_expression_data

napistu/ingestion/hpa.py

@@ -0,0 +1,147 @@
+from __future__ import annotations
+
+import logging
+import pandas as pd
+from napistu import utils
+from fs import open_fs
+from napistu.constants import ONTOLOGIES
+from napistu.ingestion.constants import PROTEINATLAS_SUBCELL_LOC_URL, PROTEINATLAS_DEFS
+
+
+logger = logging.getLogger(__name__)
+
+
+def download_hpa_data(target_uri: str, url: str = PROTEINATLAS_SUBCELL_LOC_URL) -> None:
+    """Download protein localization data from the Human Protein Atlas.
+
+    Parameters
+    ----------
+    target_uri : str
+        The URI where the HPA data should be saved. Should end with .tsv
+    url : str, optional
+        URL to download the zipped protein atlas subcellular localization tsv from.
+        Defaults to PROTEINATLAS_SUBCELL_LOC_URL.
+
+    Returns
+    -------
+    None
+
+    Notes
+    -----
+    Downloads the subcellular localization data from the Human Protein Atlas and saves
+    it to the specified target URI. The data is downloaded from the official HPA website
+    as a ZIP file and automatically unzipped to extract the TSV.
+
+    Raises
+    ------
+    ValueError
+        If target_uri does not end with .tsv
+    """
+    if not target_uri.endswith(".tsv"):
+        raise ValueError(f"Target URI must end with .tsv, got {target_uri}")
+
+    file_ext = url.split(".")[-1]
+    target_filename = url.split("/")[-1].split(f".{file_ext}")[0]
+    logger.info("Start downloading proteinatlas %s to %s", url, target_uri)
+    # target_filename is the name of the file in the zip file which will be renamed to target_uri
+    utils.download_wget(url, target_uri, target_filename=target_filename)
+
+    return None
+
+
+def load_and_clean_hpa_data(hpa_data_path: str) -> pd.DataFrame:
+    """Load and format Human Protein Atlas subcellular localization data.
+
+    Parameters
+    ----------
+    hpa_data_path : str
+        Path to HPA subcellular localization data TSV file
+
+    Returns
+    -------
+    pd.DataFrame
+        DataFrame with genes as rows and GO terms as columns. Each cell
+        is a binary value (0 or 1) indicating whether that gene (row) is found in that
+        compartment (column). Genes with no compartment annotations are filtered out.
+
+    Notes
+    -----
+    This function loads subcellular localization data from the Human Protein Atlas
+    and creates a binary matrix where rows are genes and columns are GO terms,
+    with 1 indicating that a gene is localized to that compartment and 0 indicating
+    it is not.
+
+    The function filters out genes that have no compartment annotations and logs
+    information about the number of genes filtered and the final matrix dimensions.
+
+    Raises
+    ------
+    FileNotFoundError
+        If the input file does not exist
+    ValueError
+        If no gene-compartment associations are found in the data
+    """
+    # Check file exists
+    base_path, file_name = utils.get_source_base_and_path(hpa_data_path)
+
+    logger.info("Loading Human Protein Atlas subcellular localization data")
+
+    # Read the TSV file using pandas
+    with open_fs(base_path) as base_fs:
+        with base_fs.open(file_name, "rb") as f:
+            protein_subcellular_localizations = pd.read_csv(
+                f, sep="\t", dtype=str, na_values=[""], keep_default_na=True
+            )
+
+    # Rename Gene column to be more informative
+    protein_subcellular_localizations = protein_subcellular_localizations.rename(
+        columns={PROTEINATLAS_DEFS.GENE: ONTOLOGIES.ENSEMBL_GENE}
+    )
+
+    # Convert GO id column to lists
+    def _split_go_terms(go_terms):
+        if pd.isna(go_terms):
+            return []
+        return go_terms.split(";")
+
+    # Create a list of all gene-GO term pairs
+    gene_go_pairs = []
+    for _, row in protein_subcellular_localizations.iterrows():
+        go_terms = _split_go_terms(row[PROTEINATLAS_DEFS.GO_ID])
+        for term in go_terms:
+            gene_go_pairs.append(
+                {
+                    ONTOLOGIES.ENSEMBL_GENE: row[ONTOLOGIES.ENSEMBL_GENE],
+                    ONTOLOGIES.GO: term,
+                }
+            )
+
+    # Convert to DataFrame and pivot to create binary matrix
+    gene_go_df = pd.DataFrame(gene_go_pairs)
+    if len(gene_go_df) == 0:
+        raise ValueError("No gene-compartment associations found in the data")
+
+    localization_matrix = pd.crosstab(
+        gene_go_df[ONTOLOGIES.ENSEMBL_GENE], gene_go_df[ONTOLOGIES.GO]
+    ).astype(int)
+
+    # Log number of genes without compartments that were filtered
+    n_total_genes = len(
+        protein_subcellular_localizations[ONTOLOGIES.ENSEMBL_GENE].unique()
+    )
+    n_genes_with_compartments = len(localization_matrix)
+    n_filtered = n_total_genes - n_genes_with_compartments
+    if n_filtered > 0:
+        logger.debug(
+            "Filtered out %d genes with no compartment annotations (from %d total genes)",
+            n_filtered,
+            n_total_genes,
+        )
+
+    logger.info(
+        "Created localization matrix with shape %d genes x %d compartments",
+        localization_matrix.shape[0],
+        localization_matrix.shape[1],
+    )
+
+    return localization_matrix