Build your own cell: Diffusion Models for Multichannel 3D Microscopy Image Generation

Reed Naidoo, Matt De Vries, Olga Fourkioti, Lucas G Dent, Nathan Curry, Chris Dunsby, Chris Bakal

Keywords: 3D Diffusion Models

Abstract: Three-dimensional (3D) cellular morphology is a critical indicator of cellular function, disease states, and drug responses. However, capturing and interpreting the complex relationships between cell shape, treatment conditions, and their biological implications remains a challenge. To address this, we present "Build Your Own Cell'' (BYOC), a multichannel 3D generative framework that combines vector quantisation and diffusion models to synthesise biologically realistic 3D cell structures. BYOC captures intricate morphological changes induced by different drug treatments, enabling high-throughput in silico simulations and screening of cell shapes in response to varied conditions. This novel framework represents a significant step towards accelerating pre-clinical drug development by synthesising high-resolution, biologically realistic 3D cells, potentially reducing reliance on labour-intensive experimental studies. By ensuring phenotypic consistency between cell and nucleus volumes through joint modelling, BYOC provides high-fidelity reconstructions that could facilitate downstream analyses, including drug efficacy evaluation and mechanistic studies.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12404