Add informative descriptions for all embeddings

catalog.json

@@ -28,7 +28,7 @@
     "community_contribution": false,
     "coverage_pct": 91.82,
     "dataset_path": "data/embeddings/probe_albert",
-    "description": "Protein sequence representation produced by the ALBERT protein language model and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by the ALBERT protein language model.",
     "id": "probe_albert",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_albert/metadata.json",
@@ -116,7 +116,7 @@
     "community_contribution": false,
     "coverage_pct": 91.82,
     "dataset_path": "data/embeddings/probe_bert_bfd",
-    "description": "Protein sequence representation produced by the ProtTrans BERT-BFD model and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by the ProtTrans BERT-BFD model.",
     "id": "probe_bert_bfd",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_bert_bfd/metadata.json",
@@ -138,7 +138,7 @@
     "community_contribution": false,
     "coverage_pct": 91.86,
     "dataset_path": "data/embeddings/probe_bert_pfam",
-    "description": "Protein sequence representation produced by a BERT model trained on Pfam protein-domain sequences and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by a BERT model trained on Pfam protein-domain sequences.",
     "id": "probe_bert_pfam",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_bert_pfam/metadata.json",
@@ -160,7 +160,7 @@
     "community_contribution": false,
     "coverage_pct": 95.38,
     "dataset_path": "data/embeddings/biobert_gene_protein_summary",
-    "description": "Biomedical-text representation obtained by encoding each gene's NCBI summary plus its UniProt protein summary, when available, with BioBERT and mean-pooling the text embeddings.",
+    "description": "Biomedical-text representation obtained by encoding each gene's NCBI summary plus its UniProt protein summary, when available, with BioBERT and mean-pooling the token representations.",
     "id": "biobert_gene_protein_summary",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/biobert_gene_protein_summary/metadata.json",
@@ -270,7 +270,7 @@
     "community_contribution": false,
     "coverage_pct": 93.91,
     "dataset_path": "data/embeddings/probe_blast",
-    "description": "Protein sequence-similarity representation based on BLAST-derived similarity features against a reference protein set, as distributed by PROBE.",
+    "description": "Protein sequence-similarity representation based on BLAST-derived similarity features against a reference protein set.",
     "id": "probe_blast",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_blast/metadata.json",
@@ -292,7 +292,7 @@
     "community_contribution": false,
     "coverage_pct": 91.82,
     "dataset_path": "data/embeddings/probe_cpc_prot",
-    "description": "Protein sequence representation produced by CPCProt, a recurrent self-supervised protein model, and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by CPCProt, a recurrent self-supervised protein model.",
     "id": "probe_cpc_prot",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_cpc_prot/metadata.json",
@@ -402,7 +402,7 @@
     "category_source": "manual",
     "modality": "Knowledge graph",
     "algorithm": "TransE_l2 knowledge graph embedding",
-    "description": "Knowledge-graph gene representation from pretrained TransE embeddings of the Drug Repurposing Knowledge Graph (DRKG).",
+    "description": "Knowledge-graph gene representation learned with TransE from DRKG’s heterogeneous relations among genes, compounds, diseases, biological processes, and other biomedical entities.",
     "n_genes": 18999,
     "n_dims": 400,
     "coverage_pct": 94.54,
@@ -470,7 +470,7 @@
     "community_contribution": false,
     "coverage_pct": 91.82,
     "dataset_path": "data/embeddings/probe_esmb1",
-    "description": "Protein sequence representation produced by ESM-1b and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by ESM-1b.",
     "id": "probe_esmb1",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_esmb1/metadata.json",
@@ -492,7 +492,7 @@
     "community_contribution": false,
     "coverage_pct": 80.93,
     "dataset_path": "data/embeddings/frogs_archs4_256",
-    "description": "Gene representation learned by FRoGS from large-scale ARCHS4 expression and co-expression data, capturing functional relationships among genes.",
+    "description": "Expression representation learned by FRoGS from large-scale ARCHS4 bulk RNA-seq data, capturing gene co-expression and transcriptional covariation.",
     "id": "frogs_archs4_256",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/frogs_archs4_256/metadata.json",
@@ -536,7 +536,7 @@
     "community_contribution": false,
     "coverage_pct": 82.49,
     "dataset_path": "data/embeddings/genecompass_base_gene_token",
-    "description": "Knowledge-informed gene-token representation extracted from the pretrained GeneCompass cross-species single-cell foundation model.",
+    "description": "Multimodal gene-token representation learned from large-scale single-cell transcriptomes together with promoter-sequence, gene-family, co-expression, and gene-regulatory-network priors.",
     "id": "genecompass_base_gene_token",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/genecompass_base_gene_token/metadata.json",
@@ -558,7 +558,7 @@
     "community_contribution": false,
     "coverage_pct": 99.55,
     "dataset_path": "data/embeddings/geneformer_v2_316m",
-    "description": "Static gene-token representation extracted from the input embedding table of the pretrained Geneformer V2 316M single-cell model.",
+    "description": "Static gene-token representation extracted from the Geneformer V2 316M input embedding table after pretraining on rank-encoded single-cell transcriptomes.",
     "id": "geneformer_v2_316m",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/geneformer_v2_316m/metadata.json",
@@ -648,7 +648,7 @@
     "community_contribution": false,
     "coverage_pct": 96.3,
     "dataset_path": "data/embeddings/gtex_tissue_median_log1p_zscore",
-    "description": "Expression representation based on median gene expression across GTEx tissues, transformed with log1p and standardized across genes.",
+    "description": "Expression representation based on median gene-expression profiles across GTEx tissues after log1p transformation and z-score standardization.",
     "id": "gtex_tissue_median_log1p_zscore",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/gtex_tissue_median_log1p_zscore/metadata.json",
@@ -714,7 +714,7 @@
     "community_contribution": false,
     "coverage_pct": 93.91,
     "dataset_path": "data/embeddings/probe_hmmer",
-    "description": "Protein sequence-profile representation based on HMMER-derived profile features and distributed in the PROBE benchmark.",
+    "description": "Protein sequence-profile representation based on HMMER-derived profile features.",
     "id": "probe_hmmer",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_hmmer/metadata.json",
@@ -802,7 +802,7 @@
     "community_contribution": false,
     "coverage_pct": 93.91,
     "dataset_path": "data/embeddings/probe_learned_vec",
-    "description": "Protein sequence representation learned with the unsupervised Learned-Vec method from amino-acid sequence context and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation learned with the unsupervised Learned-Vec method from amino-acid sequence context.",
     "id": "probe_learned_vec",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_learned_vec/metadata.json",
@@ -824,7 +824,7 @@
     "community_contribution": false,
     "coverage_pct": 94.13,
     "dataset_path": "data/embeddings/mahi_all_contexts_mean",
-    "description": "Context-aware Mahi gene representation obtained by averaging each gene's embeddings across the available non-global tissue and biological contexts.",
+    "description": "Multimodal gene representation obtained by averaging Mahi embeddings across tissue and cell contexts, each integrating context-specific network topology with chromatin-accessibility, transcription-factor-binding, histone-modification, and protein-structure information.",
     "id": "mahi_all_contexts_mean",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/mahi_all_contexts_mean/metadata.json",
@@ -846,7 +846,7 @@
     "community_contribution": false,
     "coverage_pct": 94.13,
     "dataset_path": "data/embeddings/mahi_global",
-    "description": "Global context-aware gene representation taken from the global Mahi embedding for each gene.",
+    "description": "Multimodal gene representation from the global Mahi context, integrating global network topology with chromatin-accessibility, transcription-factor-binding, histone-modification, and protein-structure information.",
     "id": "mahi_global",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/mahi_global/metadata.json",
@@ -934,7 +934,7 @@
     "community_contribution": false,
     "coverage_pct": 87.07,
     "dataset_path": "data/embeddings/newt_cellnet",
-    "description": "NEWT component representation derived from CellNet lineage and transcription-factor regulatory networks, encoding gene relationships in cell-identity programs.",
+    "description": "NEWT network representation derived from normalized CellNet tissue- and lineage-specific regulatory-network connectivity scores, encoding each gene’s association with lineage programs.",
     "id": "newt_cellnet",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/newt_cellnet/metadata.json",
@@ -1154,7 +1154,7 @@
     "community_contribution": false,
     "coverage_pct": 95.38,
     "dataset_path": "data/embeddings/pubmedbert_gene_protein_summary",
-    "description": "Biomedical-text representation obtained by encoding each gene's NCBI summary plus its UniProt protein summary, when available, with PubMedBERT and mean-pooling the text embeddings.",
+    "description": "Biomedical-text representation obtained by encoding each gene's NCBI summary plus its UniProt protein summary, when available, with PubMedBERT and mean-pooling the token representations.",
     "id": "pubmedbert_gene_protein_summary",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/pubmedbert_gene_protein_summary/metadata.json",
@@ -1198,7 +1198,7 @@
     "community_contribution": false,
     "coverage_pct": 94.85,
     "dataset_path": "data/embeddings/scfoundation_gene_pos",
-    "description": "Static learned gene-position representation extracted from the pretrained scFoundation single-cell foundation model.",
+    "description": "Static gene-identity representation extracted from scFoundation’s learned gene-position embedding table after pretraining on large-scale single-cell expression data.",
     "id": "scfoundation_gene_pos",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/scfoundation_gene_pos/metadata.json",
@@ -1264,7 +1264,7 @@
     "community_contribution": false,
     "coverage_pct": 99.49,
     "dataset_path": "data/embeddings/scprint_medium_v1_5_gene",
-    "description": "Learned gene-token representation extracted from the pretrained scPRINT medium v1.5 single-cell model.",
+    "description": "Learned gene-token representation extracted from scPRINT medium v1.5 after pretraining on large-scale single-cell transcriptomes.",
     "id": "scprint_medium_v1_5_gene",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/scprint_medium_v1_5_gene/metadata.json",
@@ -1286,7 +1286,7 @@
     "community_contribution": false,
     "coverage_pct": 93.91,
     "dataset_path": "data/embeddings/probe_seqvec",
-    "description": "Protein sequence representation produced by SeqVec, an ELMo-style bidirectional language model trained on protein sequences, and distributed in PROBE.",
+    "description": "Protein sequence representation produced by SeqVec, an ELMo-style bidirectional language model trained on protein sequences.",
     "id": "probe_seqvec",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_seqvec/metadata.json",
@@ -1330,7 +1330,7 @@
     "category_source": "manual",
     "modality": "Networks",
     "algorithm": "Residual graph convolutional network link prediction",
-    "description": "Network representation learned from the STRING protein-interaction graph with the GNN embedding model used for perturbation prediction.",
+    "description": "Network representation learned from the STRING protein-association network with the GNN embedding model used for perturbation prediction.",
     "n_genes": 18028,
     "n_dims": 256,
     "coverage_pct": 89.7,
@@ -1353,7 +1353,7 @@
     "category_source": "manual",
     "modality": "Networks",
     "algorithm": "WaveGC graph spectral wavelet convolution",
-    "description": "Network representation learned from the STRING protein-interaction graph with the WaveGC graph-convolution embedding model used for perturbation prediction.",
+    "description": "Network representation learned from the STRING protein-association network with the WaveGC graph-convolution embedding model used for perturbation prediction.",
     "n_genes": 18131,
     "n_dims": 128,
     "coverage_pct": 90.22,
@@ -1376,7 +1376,7 @@
     "community_contribution": false,
     "coverage_pct": 91.82,
     "dataset_path": "data/embeddings/probe_t5",
-    "description": "Protein sequence representation produced by the ProtT5 protein language model and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by the ProtT5 protein language model.",
     "id": "probe_t5",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_t5/metadata.json",
@@ -1420,7 +1420,7 @@
     "community_contribution": false,
     "coverage_pct": 94.95,
     "dataset_path": "data/embeddings/uce_33l",
-    "description": "Protein-informed gene-token representation used by the 33-layer Universal Cell Embedding (UCE) single-cell foundation model.",
+    "description": "Multimodal gene representation combining ESM2 protein-sequence information with a projection learned by UCE-33L during pretraining on large-scale single-cell transcriptomes.",
     "id": "uce_33l",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/uce_33l/metadata.json",
@@ -1442,7 +1442,7 @@
     "community_contribution": false,
     "coverage_pct": 93.91,
     "dataset_path": "data/embeddings/probe_unirep",
-    "description": "Protein sequence representation produced by UniRep, a recurrent neural network trained on large-scale protein sequence data, and distributed in PROBE.",
+    "description": "Protein sequence representation produced by UniRep, a recurrent neural network trained on large-scale protein sequence data.",
     "id": "probe_unirep",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_unirep/metadata.json",
@@ -1464,7 +1464,7 @@
     "community_contribution": false,
     "coverage_pct": 91.82,
     "dataset_path": "data/embeddings/probe_xlnet",
-    "description": "Protein sequence representation produced by a ProtTrans XLNet protein language model and distributed in the PROBE benchmark.",
+    "description": "Protein sequence representation produced by a ProtTrans XLNet protein language model.",
     "id": "probe_xlnet",
     "license": "Original source / license terms",
     "metadata_url": "https://huggingface.co/datasets/embeddingsSSRF/gene_embeddings/blob/326b257a20e73df337ece40bb82401e277851f5b/data/embeddings/probe_xlnet/metadata.json",