Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 17·5 min read
Program-space Diffusion for Morphology-to-Transcriptomics Prediction
TL;DR
A new approach to predicting gene expression from tissue morphology has been developed using a conditional diffusion model. This method leverages coordinated transcriptional variation, significantly improving the efficiency of spatial transcriptomics predictions.
✦ Why It Matters
Consider adopting this new model for more efficient and accurate gene expression predictions in your research.
Key Takeaways
- Explore the use of conditional diffusion models in your projects.
- Utilize consensus non-negative matrix factorization for data analysis.
- Focus on coordinated transcriptional variation for better predictions.
- Consider the implications of reduced dimensionality in generative tasks.
How It Works
The proposed method uses consensus non-negative matrix factorization (cNMF) to extract low-dimensional transcriptional programs from training data. These programs represent coordinated gene expression variations, which are then used to train a conditional diffusion model.
This model generates program activations based on histological images, allowing for a more integrated and efficient prediction of gene expression compared to traditional independent gene predictions.