Update data/embeddings/bioconceptvec_glove/README.md
Browse files
data/embeddings/bioconceptvec_glove/README.md
CHANGED
|
@@ -1,9 +1,7 @@
|
|
| 1 |
# Embedding name: BioConceptVec-GloVe
|
| 2 |
|
| 3 |
- Modality: Biomedical literature / text
|
| 4 |
-
- Algorithm: BioConceptVec GloVe Gene_<Entrez> concept vectors standardized by strict Entrez mapping
|
| 5 |
- Embedding source URL: https://github.com/ncbi/BioConceptVec
|
| 6 |
-
- Original model/code URL: https://github.com/ncbi/BioConceptVec
|
| 7 |
|
| 8 |
## Public files
|
| 9 |
|
|
@@ -43,15 +41,12 @@
|
|
| 43 |
|
| 44 |
This embedding represents BioConceptVec gene concepts using the GloVe training variant.
|
| 45 |
|
| 46 |
-
The raw vectors were not regenerated locally for this README update. The local work here was identifier cleaning, mapping to the shared master_gene_table.csv gene universe, duplicate handling, QC, and conversion into the shared public resource format.
|
| 47 |
-
|
| 48 |
### Provenance fields
|
| 49 |
|
| 50 |
source_name: BioConceptVec-GloVe
|
| 51 |
original_model_or_method: BioConceptVec-GloVe
|
| 52 |
input_data_type: BioConceptVec concept JSON
|
| 53 |
raw_source: ~/data/raw_embeddings/bioconceptvec/concept_glove.json
|
| 54 |
-
processing_script: ~/scripts/process_embeddings/process_bioconceptvec_glove_v1_1.py
|
| 55 |
source_download_page: https://github.com/ncbi/BioConceptVec
|
| 56 |
source_code_page: https://github.com/ncbi/BioConceptVec
|
| 57 |
source_identifier_type: Gene_<Entrez> concept key
|
|
@@ -64,7 +59,7 @@ The raw vectors were not regenerated locally for this README update. The local w
|
|
| 64 |
|
| 65 |
### Mapping and processing decisions
|
| 66 |
|
| 67 |
-
|
| 68 |
|
| 69 |
### Processing workflow
|
| 70 |
|
|
@@ -77,9 +72,6 @@ For this embedding:
|
|
| 77 |
5. Average duplicate source rows only when they map to the same final master-table gene.
|
| 78 |
6. Save the final matrix as float32.
|
| 79 |
|
| 80 |
-
### Important implementation decisions
|
| 81 |
-
|
| 82 |
-
This README describes the standardized gene-level version included in the public resource. The source vectors were kept as released or previously generated in the project, while the mapping and QC were made consistent with the rest of the v1.1 embedding library.
|
| 83 |
|
| 84 |
### Summary of final mapping:
|
| 85 |
|
|
|
|
| 1 |
# Embedding name: BioConceptVec-GloVe
|
| 2 |
|
| 3 |
- Modality: Biomedical literature / text
|
|
|
|
| 4 |
- Embedding source URL: https://github.com/ncbi/BioConceptVec
|
|
|
|
| 5 |
|
| 6 |
## Public files
|
| 7 |
|
|
|
|
| 41 |
|
| 42 |
This embedding represents BioConceptVec gene concepts using the GloVe training variant.
|
| 43 |
|
|
|
|
|
|
|
| 44 |
### Provenance fields
|
| 45 |
|
| 46 |
source_name: BioConceptVec-GloVe
|
| 47 |
original_model_or_method: BioConceptVec-GloVe
|
| 48 |
input_data_type: BioConceptVec concept JSON
|
| 49 |
raw_source: ~/data/raw_embeddings/bioconceptvec/concept_glove.json
|
|
|
|
| 50 |
source_download_page: https://github.com/ncbi/BioConceptVec
|
| 51 |
source_code_page: https://github.com/ncbi/BioConceptVec
|
| 52 |
source_identifier_type: Gene_<Entrez> concept key
|
|
|
|
| 59 |
|
| 60 |
### Mapping and processing decisions
|
| 61 |
|
| 62 |
+
Only BioConceptVec gene concepts with a single Entrez identifier were used. Non-gene concepts, multi-Entrez concepts, and ambiguous Entrez IDs were excluded. Accepted Entrez IDs were mapped through map_entrez_ids_all in the enriched master table. Duplicate source rows mapping to the same final gene were averaged.
|
| 63 |
|
| 64 |
### Processing workflow
|
| 65 |
|
|
|
|
| 72 |
5. Average duplicate source rows only when they map to the same final master-table gene.
|
| 73 |
6. Save the final matrix as float32.
|
| 74 |
|
|
|
|
|
|
|
|
|
|
| 75 |
|
| 76 |
### Summary of final mapping:
|
| 77 |
|