Update data/embeddings/depmap_crispr_gene_effect_pca256/README.md

data/embeddings/depmap_crispr_gene_effect_pca256/README.md

@@ -2,8 +2,7 @@
 
 - Modality: Functional genomics / CRISPR gene-dependency profile
 - Algorithm: PCA-reduced CRISPR gene-effect profile across cancer models
-- Embedding source URL: Local processing from DepMap CRISPRGeneEffect.csv and Model.csv
-- Original model/code URL: Not applicable; source is a DepMap CRISPR gene-effect matrix
+- Embedding source URL: CRISPRGeneEffect.csv and Model.csv from https://depmap.org/portal/download/
 
 ## Public files
 
@@ -42,7 +41,7 @@ This embedding is a gene-level functional-genomics representation derived from D
 
 The source matrix had cancer models as rows and genes as columns. Each value represents a DepMap CRISPR gene-effect score for a gene in a cancer model. During processing, the matrix was transposed to genes × models, missing values were filled by the source-gene column mean before PCA, model/cell-line features were standardized across mapped genes, and randomized PCA was applied to produce a 256-dimensional gene embedding.
 
-This embedding should therefore be interpreted as a compact CRISPR dependency-profile embedding. Genes with similar vectors have similar patterns of CRISPR dependency across DepMap cancer models. It is not a sequence embedding, expression embedding, or model-token embedding.
+This embedding should therefore be interpreted as a compact CRISPR dependency-profile embedding.
 
 ### Provenance fields
 
@@ -84,7 +83,7 @@ For the DepMap CRISPR embedding:
 
 ### Important implementation decisions
 
-PCA was used because the source representation has 1,208 model/cell-line features. The 256-dimensional PCA output captures most of the standardized dependency-profile variance while making the embedding size comparable to other compact embeddings in the resource.
+PCA was used because the source representation has 1,208 model/cell-line features.
 
 Missing values were filled before PCA because PCA requires a complete numeric matrix.