Understanding cells as integrated systems is central to modern biology. Although fluorescence microscopy can resolve subcellular structure in living cells, it is expensive, is slow, and can damage cells. We present a label-free method for predicting three-dimensional fluorescence directly from transmitted-light images and demonstrate that it can be used to generate multi-structure, integrated images. The method can also predict immunofluorescence (IF) from electron micrograph (EM) inputs, extending the potential applications.
Access optionsAccess options
Subscribe to Journal
Get full journal access for 1 year
only $18.92 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Data used to train the 3D models are available at https://downloads.allencell.org/publication-data/label-free-prediction/index.html.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank the entire Allen Institute for Cell Science team, who generated and characterized the gene-edited hiPS cell lines, developed image-based assays, and recorded the high replicate datasets suitable for modeling and without whom this work would not have been possible. We especially thank the Allen Institute for Cell Science Gene Editing, Assay Development, Microscopy, and Pipeline teams for providing cell lines and images of different transmitted-light imaging modalities, and particularly K. Gerbin, A. Nelson, and H. Malik for performing the cardiomyocyte differentiation and culture, and W. Leung, J. Tang, M. Hendershott, and N. Gaudreault for gathering the additional time series, CAAX-labeled, cardiomyocyte, HEK293, and HT-1080 data. We thank the Allen Institute for Cell Science Animated Cell team and T. Do specifically for providing expertise in figure preparation. We thank D. Fernandes for developing an early proof-of-concept 2D version of the model. We thank members of the Allen Institute for Brain Science Synapse Biology department for preparing samples and providing images that were the basis for training the conjugate array tomography data. These contributions were absolutely critical for model development. HEK293 cells were provided via the Viral Technology Laboratory at the Allen Institute for Brain Science. Cardiomyocyte and hiPSC data in this publication were derived from cells in the Allen Cell Collection, a collection of fluorescently labeled hiPSCs derived from the parental WTC line provided by B.R. Conklin, at Gladstone Institutes. We thank Google Accelerated Science for telling us about studies of 2D deep learning in neurons before we began this project. This work was supported by grants from NIH/NINDS (R01NS092474) (S.S., F.C.) and NIH/NIMH (R01MH104227) (F.C.). We thank P.G. Allen, founder of the Allen Institute for Cell Science, for vision, encouragement, and support.