CellFlow: Simulating Cellular Morphology Changes via Flow Matching

2025-02-13

AI Summary

This work addresses the critical challenge in virtual cell modeling of disentangling genuine biological effects of chemical/genetic perturbations from experimental batch noise. We propose CellFlow—the first generative framework to incorporate flow matching for modeling dynamic cellular morphology. CellFlow enables distribution-level mapping from unperturbed to perturbed states, supporting perturbation-specific generation and continuous state interpolation to effectively decouple biological signals from technical artifacts. Leveraging distribution alignment optimization and joint training on multi-source datasets (BBBC021, RxRx1, JUMP), CellFlow achieves a 35% reduction in Fréchet Inception Distance (FID) and a 12% improvement in mechanism-of-action prediction accuracy. Generated images exhibit high fidelity and biologically interpretable consistency, while enabling simulation of continuous cellular state transitions.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Web Mining and Content Analysis: Web data generation and simulation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Building a virtual cell capable of accurately simulating cellular behaviors in silico has long been a dream in computational biology. We introduce CellFlow, an image-generative model that simulates cellular morphology changes induced by chemical and genetic perturbations using flow matching. Unlike prior methods, CellFlow models distribution-wise transformations from unperturbed to perturbed cell states, effectively distinguishing actual perturbation effects from experimental artifacts such as batch effects -- a major challenge in biological data. Evaluated on chemical (BBBC021), genetic (RxRx1), and combined perturbation (JUMP) datasets, CellFlow generates biologically meaningful cell images that faithfully capture perturbation-specific morphological changes, achieving a 35% improvement in FID scores and a 12% increase in mode-of-action prediction accuracy over existing methods. Additionally, CellFlow enables continuous interpolation between cellular states, providing a potential tool for studying perturbation dynamics. These capabilities mark a significant step toward realizing virtual cell modeling for biomedical research.

Problem

• Simulating cellular morphology changes
• Distinguishing perturbation effects from artifacts
• Improving image generation and prediction accuracy

Innovation

• Flow matching for cell simulation
• Distinguishes perturbation effects accurately
• Enables continuous cell state interpolation