Competition Overview
847
Active Teams
12.4K
Submissions
0.847
Best Corr
0.763
Mean Corr
3.2M
Cells
458
Perturbations
Model Performance
Pearson Correlation
Competition Day
Baseline: 0.621 Current: 0.847
Top model Pearson correlation over time. Measures how well predicted gene expression matches actual post-perturbation expression across all genes.
Challenge Timeline
Jan 15
Registration Opens Completed
Feb 1
Training Data Release Active
Apr 15
Validation Phase 43 days
May 30
Final Submissions 89 days
Leaderboard
Live Updated 2 min ago
Pearson correlation measures linear relationship between predicted and actual gene expression (0.0 = no correlation, 1.0 = perfect). Scores above 0.80 indicate strong predictive performance.
Public Leaderboard
Private Test
Reproducibility
Rank
Team
Pearson Corr
Spearman Corr
MSE (log)
Submissions
1
DeepCell Dynamics
0.8472
0.8156
0.1342
247
2
BioTransformer Lab
0.8398
0.8089
0.1467
189
3
Neural Perturbation
0.8245
0.7912
0.1598
312
4
CellVAE Collective
0.8187
0.7867
0.1712
156
5
Genomic Gradient
0.8023
0.7734
0.1845
278
Submit Model
Drag & drop your model checkpoint
Supports .pt, .h5, .ckpt formats (max 2GB)
REQUIREMENTS
  • ✓ Model config JSON
  • ✓ Inference script
  • ✓ Requirements.txt
  • ✓ Docker container (optional)
Cell Embeddings Visualization Live Preview
UMAP 1
UMAP 2
T Cells
B Cells
Monocytes
NK Cells
UMAP (Uniform Manifold Approximation) reduces high-dimensional gene expression data to 2D for visualization. Clusters indicate similar cell types based on RNA expression patterns.
Community Hub
Discord Community
3,847 members online
RECENT DISCUSSIONS
@neural_net: New augmentation strategy improved F1 by 2.3%
@bioml_lab: Anyone tried contrastive learning on perturbation data?
@celldynamics: Sharing pretrained embeddings for PBMC dataset
RESOURCES
Dataset Overview
COMPETITION TRACKS
RNA-seq Track: Single-cell RNA sequencing data measuring gene expression across thousands of genes per cell. Models predict genome-wide gene expression changes in response to chemical/genetic perturbations. Evaluated on held-out perturbations not seen during training.
Coming 2026: Multimodal track combining RNA + protein expression (CITE-seq)
TRAINING SET
2.8M
Single Cells
347
Perturbations
18,752
Genes
VALIDATION SET
0.4M
Single Cells
89
Perturbations
18,752
Genes
TEST SET (HIDDEN)
???
Single Cells
???
Perturbations
18,752
Genes