Elucidating the Design Space of Flow Matching for Cellular Microscopy

arXiv cs.CV / 3/31/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper examines and maps the underexplored design space of flow-matching generative models specifically for cellular microscopy image simulation.
It finds that several commonly used techniques for flow-matching are unnecessary and may degrade performance, providing guidance on what to omit or change in model construction.
The authors propose a “simple, stable, and scalable” training recipe and use it to train a foundation model for cell-response simulation.
Scaling the approach to roughly two orders of magnitude larger than prior work yields substantially improved image quality metrics (two-fold FID and ten-fold KID).
By fine-tuning with pre-trained molecular embeddings, the model achieves state-of-the-art performance for simulating responses to previously unseen molecules.

Abstract

Flow-matching generative models are increasingly used to simulate cell responses to biological perturbations. However, the design space for building such models is large and underexplored. We systematically analyse the design space of flow matching models for cell-microscopy images, finding that many popular techniques are unnecessary and can even hurt performance. We develop a simple, stable, and scalable recipe which we use to train our foundation model. We scale our model to two orders of magnitude larger than prior methods, achieving a two-fold FID and ten-fold KID improvement over prior methods. We then fine-tune our model with pre-trained molecular embeddings to achieve state-of-the-art performance simulating responses to unseen molecules. Code is available at https://github.com/valence-labs/microscopy-flow-matching