Generative semantic multiplexing (SemaPlex) for accessible and scalable multiplexed fluorescence imaging

Abstract

Multiplexed fluorescence imaging enhances spatially-resolved interrogation of complex, multi-molecular cell processes that are insufficiently sampled using standard 4-5 plex imaging. To improve accessibility and scalability for multiplexed imaging, we demonstrate generative Semantic Multiplexing (SemaPlex); a simple experimental and deep learning strategy for amplifying marker plexity several-fold by semantically unmixing multiple markers combined per imaging channel. We first characterise key determinants of SemaPlex performance, achieving precise computational multiplexing of 2-to-8 markers synthetically mixed in one channel, facilitating enhanced cell phenotype classification. We then demonstrate practical SemaPlex application, acquiring 10 markers over 4 channels (3*3-plex+1) to efficiently emulate real multiplexed labelling. This permitted accurate reconstruction of quantitative single-cell phenotypic manifolds delineating cell-cycle and mitotic dynamics, with internally validated error-detection. Finally, we exemplify use of semantic guides; additional input channels that significantly enhance multiplexing fidelity. SemaPlex makes multiple-fold increases in fluorescence imaging-plexity accessible, scalable and customisable; democratising multiplexed imaging-based interrogation of complex cell biology.