biscot: an Optimal Transport framework for multimodal bacterial single-cell data analysis

Computational optimal transport-based approaches have emerged as promising tools for the integration and interpretation of complex single-cell data. In this study, we introduce an integrative Optimal Transport (OT) framework for spatiotemporal and multi-omics bacterial single-cell analysis using Gaussian Mixture Model (GMM) OT, termed biscot (bacterial integrative single-cell optimal transport). We show that biscot, equipped with a novel global-to-local GMM initialization, outperforms classical OT and entropically-regularized OT methods both in terms of speed and accuracy for disentangling complex bacterial communities mixtures from single-cell flow cytometry data. When applied to time-series flow cytometry data from Bacillus subtilis , our framework delivers robust and biologically meaningful results, effectively capturing subtle phenotypic shifts in spore populations transitioning from inactive to active growth states. biscot also allows multi-omics integration of flow cytometry and unpaired bacterial single-cell RNA sequencing (scRNA-seq) data, enabling the alignment of individual gene expression profiles to the cytometric data. For an unpaired flow cytometry/scRNA-seq dataset of Bacillus subtilis cells, we validate the biological plausibility of inferred gene expression patterns with relevant marker genes, including spoVID and nin , closely aligning with observed cellular states. Overall, our framework thus provides not only dynamic tracking of phenotypic cell states but aligns cell states with detailed transcriptomic information from scRNA-seq, demonstrating its potential to advance microbial single-cell research. biscot will be made publicly available on GitHub.