Update README.md

README.md

@@ -18,11 +18,6 @@ The first comprehensive genome-wide virtual screening dataset for human proteins
 ### Dataset Description  
 **GenomeScreen** represents the largest and most comprehensive virtual screening dataset ever created, containing over **2 million potential drug hits** targeting **>20,000 binding pockets** across **~10,000 human proteins** predicted by AlphaFold2. This dataset was generated through an unprecedented **10 trillion protein-ligand scoring operations**, covering nearly half of the human genome.
 
-Key features:
-- Screened **500+ million drug-like molecules** from ZINC and Enamine REAL databases
-- Completed in just **24 hours** on 8×A100 GPUs using DrugCLIP technology
-- Includes **diverse chemical clusters** selected via CADD pipeline
-- Covers **all major human protein families**
 
 - **Curated by:** Yinjun Jia, Bowen Gao, Jiaxin Tan, Jiqing Zheng, Xin Hong, et al. (Tsinghua University)
 - **Hosted at:** [Drug-The-Whole-Genome Portal](https://drug-the-whole-genome.yanyanlan.com)
@@ -36,53 +31,24 @@ Key features:
 
 ---
 
-## Uses  
-
-### Direct Use  
-- **Hit Identification:** Rapid discovery of lead compounds for novel targets
-- **Target Druggability:** Assess ligandability of human proteins at scale
-- **Method Benchmarking:** Test docking/scoring algorithms against gold-standard data
-- **AI Training:** Train foundation models for structure-based drug design
-
-### Out-of-Scope Use  
-- **Clinical Decision Making:** Requires experimental validation
-- **Absolute Affinity Prediction:** Scores reflect relative rankings only
-- **Non-human Applications:** Focused exclusively on human proteome
-
----
-
 ## Dataset Structure  
 
 ### Files and Formats  
 ```
-|- target_protein/
-  |- pocket_1/
-    |- refined_structure.pdb.gz # Processed AlphaFold2 structure
-    |- docking_grid.in # Pre-computed grid coordinates
-    |- top_molecules.csv # Ranked hits with DrugCLIP scores
-    |- clustered_hits/ # Chemically diverse subsets
-    |- cluster_representatives.csv
-  |- pocket_2/
+|- folder of target_domain_pocket
+    |- target_domain_pocket_template/conformer.pdbgz  # Refined protein structure based on aligned template or gen-packed conformer
+    |- xxx_grid.in  # Grid center coordinates for docking (specific to the pocket)
+    |- leader.csv  # Representative molecules selected from clusters of top 10K ranked by DrugCLIP
 ```
 
 
-### Key Contents  
-- **Protein Structures:** Processed AlphaFold2 predictions with defined pockets
-- **Docking Grids:** Pre-computed binding site coordinates (x,y,z)
-- **Ranked Hits:** Top molecules with SMILES, DrugCLIP scores, and poses
-- **Cluster Representatives:** Diverse subsets for hit expansion
-
----
-
 ### Statistics  
 | Metric                     | Value               |
 |----------------------------|---------------------|
 | Proteins Screened          | ~10,000            |
 | Binding Pockets            | >20,000            |
-| Molecules Screened          | 500+ million       |
-| Scoring Operations          | >10 trillion       |
-| Potential Hits Identified  | 2+ million         |
-| GPU Hours                  | 192 (8×A100×24h)   |
+| Molecules Screened         | >500 million       |
+| Potential Hits Identified  | >2 million         |
 
 ---