Researchers have developed scVision, a novel vision foundation model for single-cell biology that represents cells as continuous images rather than gene tokens. This approach leverages spatial gene cartography, arranging co-expressed genes as spatial neighbors to form an image where gene programs appear as local textures. Pre-trained on millions of human cells using a vision transformer and masked image modeling, scVision demonstrates superior accuracy in zero-shot cell-type annotation and gene program recovery compared to existing models and classical baselines. The model's performance is significantly influenced by the biological meaningfulness of gene positioning, highlighting the effectiveness of reframing single-cell representation learning as a computer vision problem. AI
IMPACT This novel approach could significantly advance biological research by enabling more accurate and insightful analysis of single-cell data through advanced computer vision techniques.
RANK_REASON The cluster contains an academic paper detailing a new model and methodology for biological analysis. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Hugging Face
- Human cells lacking mtDNA: repopulation with exogenous mitochondria by complementation
- masked image modelling
- optimal transport
- Ridvan Yesiloglu
- scVision
- vision transformer
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
