tools
OpenPhenom-S/16
OpenPhenom-S/16 turns microscopy images into 384-dimensional embeddings for biological comparison and downstream modeling.
OpenPhenom-S/16 is a channel-agnostic Vision Transformer for microscopy data. It accepts variable numbers of imaging channels in any order and produces embeddings that can support classification, similarity analysis, and studies of genetic or chemical perturbations.
Researchers can access it through Google Cloud Vertex AI Model Garden or Hugging Face. The model is available for non-commercial use; commercial use requires contacting Recursion. It is not a complete analysis workflow or drug-discovery platform by itself.
Features
- Processes microscopy images with variable numbers of channels
- Accepts imaging channels in any order
- Produces 384-dimensional image embeddings
- Supports one joint embedding for each input image
- Can optionally return an embedding for each input channel
- Supports downstream classification and similarity modeling
- Available through Vertex AI Model Garden and Hugging Face
Use cases
- Compare microscopy images using cosine similarity
- Classify cellular or tissue image features
- Relate genetic perturbations in high-dimensional embedding space
- Study compound-gene interactions without task-specific training
- Transfer embeddings to other fluorescence microscopy datasets
- Apply image representations to histology research