Update README.md

README.md

@@ -21,9 +21,9 @@ size_categories:
 
 HVUE (Human Virome Understanding Evaluation) is a comprehensive benchmark for evaluating foundation models on viral genomics tasks. The benchmark comprises 7 curated datasets across 3 epidemiologically critical prediction tasks:
 
-- **Pathogenicity Classification** (3 datasets, ~50,000 sequences)
-- **Host Tropism Prediction** (1 dataset, 9,428 sequences)
-- **Transmissibility Assessment** (3 datasets, ~7,300 sequences)
+- **Pathogenicity Classification** (3 datasets)
+- **Host Tropism Prediction** (1 dataset)
+- **Transmissibility Assessment** (3 datasets)
 
 **Paper**: *HViLM: A Foundation Model for Viral Genomics Enables Multi-Task Prediction of Pathogenicity, Transmissibility, and Host Tropism*
 
@@ -72,9 +72,9 @@ HVUE (Human Virome Understanding Evaluation) is a comprehensive benchmark for ev
 ## Data Format
 
 Each dataset contains three splits:
-- `train.csv` (70%)
-- `dev.csv` (15%)
-- `test.csv` (15%)
+- `train.csv` 
+- `dev.csv` 
+- `test.csv` 
 
 CSV columns:
 - `sequence`: Viral genomic sequence (250-1000 bp)
@@ -94,12 +94,7 @@ patho_cini = load_dataset("duttaprat/HVUE", data_files="pathogenicity/CINI/*.csv
 train_data = load_dataset("duttaprat/HVUE", data_files="pathogenicity/CINI/train.csv")
 ```
 
-## Data Leakage Prevention
 
-Systematic overlap analysis was performed between pre-training data (VIRION database) and HVUE benchmark:
-- Method: Accession ID matching + MMseqs2 similarity (>95% identity)
-- Result: 186 overlapping sequences removed from pre-training corpus
-- Clean separation between pre-training and evaluation data
 
 ## Citation
 ```bibtex