Shorten preparation section in dataset card

README.md

@@ -1,12 +1,30 @@
 ---
-viewer: true
+viewer: false
 ---
 
 # OAS95-aligned-cleaned
 
-Minimal dataset card.
+A cleaned and IMGT-aligned antibody sequence dataset derived from the Observed Antibody Space (OAS) database.
 
-Exact row counts:
-- train: 203968932
-- test: 11819793
-- total: 215788725
+## Preparation
+
+- Started from the OAS dataset preprocessed with the IgLM pipeline and clustered at 95% sequence identity.
+- Numbered all sequences with ANARCII using the IMGT numbering scheme.
+- Removed heavy- and light-chain sequences with numbering gaps at the first framework position.
+- Removed light-chain sequences shorter than 90 residues.
+- Retained 203,968,932 training sequences and 11,819,793 test sequences across 6 organism labels and 2 chain types.
+
+## Columns
+
+| Column | Description |
+| --- | --- |
+| `sequence` | IMGT-aligned amino acid sequence with gap characters. |
+| `init_seq` | Original amino acid sequence without alignment gaps. |
+| `class` | Organism label. |
+| `type` | Antibody chain type: `Heavy` or `Light`. |
+
+## Split Sizes
+
+- train: 203,968,932
+- test: 11,819,793
+- total: 215,788,725