Embedding name: Mashup-STRING

Modality: PPI / network
Algorithm: Pretrained Mashup STRING human network vectors standardized by enriched symbol/protein-ID mapping
Embedding source URL: https://mashup.csail.mit.edu/

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: 16,331 genes × 800 dimensions
Dtype: float32

Detailed processing notes

Source and interpretation

This embedding represents genes using the pretrained Mashup human STRING network vectors.

Mashup is a network-integration method. In this resource, the retained matrix is not a newly trained Mashup model. It is the standardized gene-level version of pretrained Mashup/STRING human vectors, mapped into the shared master_gene_table.csv universe.

Provenance fields

source_name: Mashup-STRING
original_model_or_method: Mashup-STRING
input_data_type: pretrained Mashup STRING human vectors and matching gene list
raw_source: ~/data/raw_embeddings/mashup/
source_download_page: https://mashup.csail.mit.edu/
source_code_page: https://mashup.csail.mit.edu/
source_identifier_type: Mashup gene name / gene symbol, with some source records requiring protein-ID rescue
mapping_identifier_type: priority symbol mapping plus enriched protein-ID mapping to master_gene_table.csv
source_rows_or_entries: 18,362 raw source rows
final_mapped_genes: 16,331
model_or_feature_dimension: 800
training_or_finetuning: none; source vectors were not retrained locally
final_embedding_dtype: float32

Source-processing decisions

The source was treated as an already-computed network embedding. Therefore, the local processing did not rebuild STRING networks and did not retrain Mashup.

Mapping and duplicate-handling decisions

Mashup source identifiers were mapped mainly through gene-symbol matching. The mapping priority was current master-table gene symbol / HGNC-approved symbol first, then HGNC previous symbols, then HGNC aliases. Source protein-like identifiers were also mapped through enriched protein-ID columns when available.

Only unique mappings were accepted. Source identifiers that could map to more than one master-table gene were excluded. When multiple accepted source rows mapped to the same final master-table gene, their vectors were averaged to keep one final vector per gene.

Processing workflow

For this embedding:

1. Read the pretrained Mashup human STRING vector table and its matching source gene list.
2. Align each source row with its source identifier.
3. Normalize source identifiers for symbol/protein-ID mapping.
4. Map source identifiers to master_gene_table.csv using the enriched symbol and protein-ID fields.
5. Exclude unmapped and ambiguous source identifiers.
6. Average duplicate source rows only when they map to the same final master-table gene.
7. Save the final gene-aligned matrix as float32.

Important implementation decisions

The central requirement was that every retained row must align to one gene in master_gene_table.csv and that the matrix row order must match genes.tsv exactly.

For Mashup, the original biological representation was preserved from the downloaded pretrained Mashup/STRING vectors.

Summary of final mapping:

Source identifiers / source rows or entries: 18,362 raw source rows
Dimensions: 800
Final mapped genes: 16,331
Master table rows: 20,097
Master coverage: 81.26%
Mapping strategy: priority symbol mapping plus enriched protein-ID mapping to master_gene_table.csv
Ambiguous identifiers: excluded
Duplicate final mappings averaged: yes

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`
`mapping_strategy`	`mapping_method`
`num_genes`	`n_genes`
`embedding_dimension`	`n_dimensions`
`matrix_shape[0]`	`n_genes`
`matrix_shape[1]`	`n_dimensions`
`dtype_saved`	`dtype`

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

{
  "all_zero_rows": {
    "count": 0,
    "documented": true,
    "expected": false,
    "reason": "No exact all-zero rows were detected in embeddings.npz.",
    "report_file": null
  },
  "ambiguous_source_rows": "excluded and reported",
  "coverage_of_master_rows": 0.8126088470916057,
  "duplicate_source_rows": "averaged when multiple source rows mapped to the same master gene",
  "embedding_dim": 800,
  "former_processing_embedding_id": "mashup_string_v1_1",
  "mapping_method_counts": {
    "direct_canonical_protein_id": 17,
    "direct_current_gene_symbol": 15037,
    "ensembl_grch37_gtf_protein_id": 487,
    "ensembl_grch38_current_gtf_protein_id": 6,
    "hgnc_alias_symbol": 1,
    "hgnc_previous_symbol": 811
  },
  "n_ambiguous_source_rows_excluded": 23,
  "n_mapped_genes": 16331,
  "n_mapped_source_rows": 16359,
  "n_master_rows": 20097,
  "n_nonfinite_source_rows_skipped": 0,
  "n_source_ensp_like_rows": 2352,
  "n_source_rows_raw": 18362,
  "n_source_symbol_like_rows": 16010,
  "n_unmapped_source_rows": 1980,
  "n_zero_vectors": 0,
  "public_file_structure": {
    "embeddings": "embeddings.npz",
    "genes": "genes.tsv",
    "metadata": "metadata.json",
    "readme": "README.md"
  }
}