Understanding disease biology at the subcellular level is key to precision medicine. This technology introduces NOVA – Neuronal Organellomics Vision Atlas, an AI-powered phenomics platform that analyzes millions of confocal images to quantify structural and spatial changes across 25 cellular organelles. Using vision transformers and contrastive learning, NOVA converts microscopic images into quantitative cellular signatures, enabling high-content, unbiased phenotyping for drug discovery and disease modeling.
- High-content phenotypic screening of therapeutic compounds
- Mechanistic insights into drug-induced subcellular effects
- Discovery of disease-specific organellar biomarkers
- Patient-derived cell phenotyping for diagnostic and prognostic use, integrable with other omics datasets for target validation
- Detects subtle, multi-organelle changes
- Single-cell resolution: Maps membrane-bound and membrane-less organelles simultaneously
- Unbiased analysis: Segmentation-free approach generalizable across cell types
- Disease relevance: Accurately distinguishes ALS patient neurons from controls
Vision transformer–based perturbation learning maps neuronal organellomics.
(A) The 25 organelles analyzed.
(B) Illustration of the perturbation learning workflow.
(C) Integrated organellome characterization of human ALS neurons based on 4 organelles in 60-day iPSC-derived motoneurons from controls and ALS patients
The NOVA model has been trained on >3 million neuronal images from iPSC-derived neurons and validated on multiple cell types and perturbations. Proof-of-concept studies identified new ALS-related organellar mechanisms and robust disease classification performance. The platform can be readily extended to additional disease models and compound screens.
