Update data/embeddings/uce_33l/README.md
Browse files
data/embeddings/uce_33l/README.md
CHANGED
|
@@ -2,8 +2,7 @@
|
|
| 2 |
|
| 3 |
- Modality: Single-cell expression foundation model / protein-informed gene-token embedding
|
| 4 |
- Algorithm: UCE fixed gene-token vector from official token table and encoder projection
|
| 5 |
-
- Embedding source URL: https://github.com/snap-stanford/UCE
|
| 6 |
-
- Original model/code URL: https://github.com/snap-stanford/UCE
|
| 7 |
|
| 8 |
## Public files
|
| 9 |
|
|
@@ -48,7 +47,7 @@ This embedding is a fixed gene-level representation derived from the official Un
|
|
| 48 |
|
| 49 |
The retained representation uses the official human token segment from the UCE token resources and the UCE encoder projection to generate fixed 1,280-dimensional gene vectors. Contextual transformer layers were not used because those outputs depend on the input cell context and would not define a single reproducible vector per gene.
|
| 50 |
|
| 51 |
-
This embedding should therefore be interpreted as a protein-informed UCE gene-token representation
|
| 52 |
|
| 53 |
### Provenance fields
|
| 54 |
|
|
|
|
| 2 |
|
| 3 |
- Modality: Single-cell expression foundation model / protein-informed gene-token embedding
|
| 4 |
- Algorithm: UCE fixed gene-token vector from official token table and encoder projection
|
| 5 |
+
- Embedding source and original model URL: https://github.com/snap-stanford/UCE
|
|
|
|
| 6 |
|
| 7 |
## Public files
|
| 8 |
|
|
|
|
| 47 |
|
| 48 |
The retained representation uses the official human token segment from the UCE token resources and the UCE encoder projection to generate fixed 1,280-dimensional gene vectors. Contextual transformer layers were not used because those outputs depend on the input cell context and would not define a single reproducible vector per gene.
|
| 49 |
|
| 50 |
+
This embedding should therefore be interpreted as a protein-informed UCE gene-token representation.
|
| 51 |
|
| 52 |
### Provenance fields
|
| 53 |
|