Agentic data science with the Data Agent Kit
cloud.google.com·8h ago
AI Research
arXiv CS.AI·Aug 25·5 min read
DeMixPert: Decomposed Response Modeling with Gaussian Mixtures for OOD Single-Cell Perturbation Prediction
TL;DR
DeMixPert is a novel approach for predicting single-cell responses to unseen genetic perturbations by decomposing transcriptional changes into systematic and perturbation-specific components. It outperforms existing methods by effectively capturing heterogeneous responses across various perturbations.
✦ Why It Matters
Implement DeMixPert to enhance your predictive models for single-cell genetic perturbations today.
Key Takeaways
- Utilize DeMixPert for improved prediction of genetic perturbation responses.
- Incorporate Gaussian prototypes for adaptive modeling of population-level variation.
- Leverage pretrained embeddings for better generalization to unseen targets.
- Access the source code to integrate DeMixPert into your research.
How It Works
DeMixPert decomposes the response to genetic perturbations into three distinct components: a systematic response based on the basal state of control cells, a perturbation-specific response inferred from pretrained target embeddings, and population-level variation modeled through a Gaussian prototype Invertible Network. This allows the model to adaptively combine Gaussian prototypes based on the specific basal state and perturbation conditions, resulting in a more accurate representation of the variation distribution.