The Light My Cells dataset, developed through the first France-BioImaging challenge, has been published in Nature Scientific Data ⁽¹⁾. It represents the first scientific outcome of this initiative which involved 31 contributors across the France-BioImaging network from Montpellier, Toulouse, Marseille, Paris, Rennes, Bordeaux and Strasbourg.

Predicting fluorescence from label-free imaging

Fluorescence microscopy is widely used across life and physical sciences to access specific molecular or structural information within samples. However, it is constrained by phototoxicity, photobleaching, and demanding sample preparation, which can limit its use in long-term observations and high-throughput experiments.

In contrast, transmitted light techniques, such as bright-field, phase contrast, and differential interference contrast, enable non-invasive imaging without labeling, preserving cell integrity over time. However, they do not provide direct molecular specificity.

The Light My Cells challenge builds on existing approaches and stimulates the development of deep learning methods to predict fluorescence signals from transmitted light images, opening new perspectives for label-free and less invasive imaging strategies

A large-scale dataset built across France-BioImaging

The publication introduces an open-access dataset developed within the France-BioImaging infrastructure.

It brings together 2,574 acquisition sets (unique fields of view), corresponding to a total of 56,984 two-dimensional microscopy images, derived from 30 independent studies collected across 8 imaging centers within the France-BioImaging infrastructure.

Each acquisition set combines transmitted light images with at least one fluorescence image targeting key subcellular structures, including the nucleus, mitochondria, tubulin, and actin. The diversity of samples, imaging systems, and experimental conditions supports the development of robust and generalizable models.

To ensure interoperability and reuse, all data were standardized using the OME-TIFF format and enriched with REMBI-compliant metadata, following FAIR data principles. A dedicated preprocessing pipeline further ensures consistency, including best-focus selection and harmonized data structure.

First scientific outcome of the Light My Cells challenge

The dataset is publicly available through the BioImage Archive, providing a reliable resource for the scientific community.

This publication represents the first scientific outcome of the Light My Cells challenge, providing a structured and openly accessible dataset designed to support the development and evaluation of deep learning models for fluorescence prediction from transmitted light microscopy.

⁽¹⁾Kauffmann, D., Gay, G., Mateos-Langerak, J. et al. 2D Multimodal Image Collection for Fluorescence Prediction from Transmitted Light Microscopy. Sci Data (2026). https://doi.org/10.1038/s41597-026-07004-w