Blame for data/embeddings/probe_learned_vec/README.md

# Embedding name: PROBE-LEARNED-VEC

- Modality: Protein representation
- Embedding source URL: https://huggingface.co/spaces/HUBioDataLab/PROBE
- Original model/code URL: https://github.com/HUBioDataLab/PROBE

## Public files

- `embeddings.npz`: compressed NumPy archive containing the embedding matrix under key `embeddings`.
- `genes.tsv`: row metadata table. Row order matches the embedding matrix.
- `metadata.json`: machine-readable provenance and processing metadata.

## Matrix

- Shape: `18,873 genes × 64 dimensions`
- Dtype: `float32`



## Detailed processing notes

### Source and interpretation

This embedding is one PROBE human UniProt representation table standardized into the shared gene-embedding resource.

This is a learned protein representation from the PROBE archive.

The source file used for this component was `LEARNED-VEC_UNIPROT_HUMAN.csv` from the official PROBE human UniProt representation archive. 

### Provenance fields

    source_name: PROBE-LEARNED-VEC
    original_model_or_method: PROBE human UniProt representation archive
    input_data_type: official PROBE human UniProt CSV representation table
    input_source_file: LEARNED-VEC_UNIPROT_HUMAN.csv
    source_download_or_app_page: https://huggingface.co/spaces/HUBioDataLab/PROBE
    source_code_page: https://github.com/HUBioDataLab/PROBE
    raw_source_folder: ~/data/raw_embeddings/probe_official/human_representation_vectors/revision-1/
    source_identifier_type: UniProt Entry accession
    mapping_identifier_type: UniProt accession matched to enriched master_gene_table.csv UniProt fields
    source_rows_or_entries: 20,421 raw rows
    final_mapped_genes: 18,873
    model_or_feature_dimension: 64
    training_or_finetuning: none
    final_embedding_dtype: float32

### Mapping and processing decisions

Each PROBE CSV uses a UniProt accession in the `Entry` column. Source entries were cleaned to their canonical accession base by removing isoform suffixes when present.

Cleaned UniProt accessions were mapped to master_gene_table.csv using `uniprot_id`, `hgnc_uniprot_ids`, and `map_uniprot_ids_all`. Only unique UniProt-to-master mappings were accepted. Ambiguous UniProt accessions were excluded. Duplicate source proteins mapping to the same final master-table gene were averaged within that gene.

### Processing workflow

For this PROBE embedding:

    1. Read `LEARNED-VEC_UNIPROT_HUMAN.csv` from the PROBE human representation archive.
    2. Use the `Entry` column as the source UniProt accession.
    3. Remove isoform suffixes where present to obtain the canonical UniProt accession base.
    4. Map source UniProt accessions to master_gene_table.csv through enriched UniProt fields.
    5. Exclude unmapped and ambiguous UniProt accessions.
    6. Average duplicate source rows only when they map to the same final master-table gene.
    7. Save the final matrix as float32.

### Important implementation decisions

All PROBE embeddings use the same mapping logic, so differences between PROBE outputs reflect the source representation type rather than different gene-mapping rules.

### Summary of final mapping:

    Source identifiers / source rows or entries: 20,421 raw rows
    Dimensions: 64
    Final mapped genes: 18,873
    Master table rows: 20,097
    Master coverage: 93.91%
    Mapping strategy: UniProt Entry accession mapping through enriched master table
    Ambiguous UniProt accessions: excluded
    Duplicate final mappings averaged: yes

<!-- BEGIN CATALOG NORMALIZATION: LEGACY METADATA -->

## Processing and provenance metadata

This section preserves information moved out of `metadata.json` during catalog normalization. The loading-facing metadata is intentionally compact; detailed source, mapping, processing, quality-control, and reproducibility records remain here.

Source catalog commit: `ab37742040d20b523aebd188d3079871a0c2b534`.

### Canonicalized field names

| Previous field | Compact metadata field |
| --- | --- |
| `embedding_name` | `display_name` |
| `source_identifier_type` | `original_identifier_type` |
| `mapping_strategy` | `mapping_method` |
| `num_genes` | `n_genes` |
| `embedding_dimension` | `n_dimensions` |
| `matrix_shape[0]` | `n_genes` |
| `matrix_shape[1]` | `n_dimensions` |
| `original_model_or_method` | `original_method` |
| `download_source` | `input_data_source` |

### Derived standardized fields

- `species` was set to `Homo sapiens` from the validated human Ensembl `ENSG` identifier namespace.
- `master_coverage_fraction` was computed as `n_genes / 20097`.

### Legacy processing and provenance record

```json
{
  "all_zero_rows": {
    "count": 0,
    "documented": true,
    "expected": false,
    "reason": "No exact all-zero rows were detected in embeddings.npz.",
    "report_file": null
  },
  "counts": {
    "ambiguous_source_rows": 186,
    "duplicate_final_mappings_averaged": 30,
    "final_dimensions": 64,
    "final_mapped_genes": 18873,
    "mapped_source_rows_before_duplicate_collapse": 18905,
    "master_rows": 20097,
    "source_dimensions": 64,
    "source_rows": 20421,
    "unmapped_source_rows": 1330
  },
  "coverage": {
    "master_gene_coverage_fraction": 0.9390953873712494,
    "source_row_mapping_fraction": 0.9257626952646785
  },
  "former_processing_embedding_id": "probe_learned_vec_v1_1",
  "mapping_method_counts": {
    "uniprot_accession_exact_unique": 18873
  },
  "public_file_structure": {
    "embeddings": "embeddings.npz",
    "genes": "genes.tsv",
    "metadata": "metadata.json",
    "readme": "README.md"
  },
  "sequence_and_reference_provenance": {
    "input_sequence_source": "PROBE human UniProt representation archive",
    "input_sequence_type": "protein sequence-derived representation",
    "sequence_reference_status": "PROBE archive provenance recorded; exact UniProt release to verify from original PROBE documentation if needed",
    "source_identifier": "UniProt accession in Entry column",
    "species": "human"
  },
  "source_embedding": "PROBE_LEARNED-VEC"
}
```

<!-- END CATALOG NORMALIZATION: LEGACY METADATA -->