Learning single-cell perturbation responses using neural optimal transport.

Understanding and predicting molecular responses in single cells upon chemical, genetic or mechanical perturbations is a core question in biology. Obtaining single-cell measurements typically requires the cells to be destroyed. This makes learning heterogeneous perturbation responses challenging as we only observe unpaired distributions of perturbed or non-perturbed cells. Here we leverage the theory of optimal transport and the recent advent of input convex neural architectures to present CellOT, a framework for learning the response of individual cells to a given perturbation by mapping these unpaired distributions. CellOT outperforms current methods at predicting single-cell drug responses, as profiled by scRNA-seq and a multiplexed protein-imaging technology. Further, we illustrate that CellOT generalizes well on unseen settings by (1) predicting the scRNA-seq responses of holdout patients with lupus exposed to interferon-ß and patients with glioblastoma to panobinostat; (2) inferring lipopolysaccharide responses across different species; and (3) modeling the hematopoietic developmental trajectories of different subpopulations.

Iterative Indirect Immunofluorescence Imaging (4i) on Adherent Cells and Tissue Sections.

Highly multiplexed protein measurements from multiple spatial scales using fluorescence microscopy recently emerged as a powerful way to investigate tumor microenvironments in biomedicine and the multivariate nature of complex systems' interactions. A range of methods for this exist, which either rely on directly labeling the primary antibody with oligonucleotides/rare metals or employing methods to remove fluorescence for cyclic acquisition. Here, we describe a protocol that uses off-the-shelf primary and secondary antibodies without further need for modification and only commonly available chemical reagents. The method harnesses the observation that antibodies can crosslink to bound epitopes during light exposure, thus preventing elution. By utilizing a simple oxygen radical scavenging buffer during imaging and by blocking free sulfhydryl groups before antibody incubation, the presented method can employ comparably mild conditions to remove bound antibodies from epitopes, which preserves sample integrity. Thus, with the stated minor modifications, it allows for a standard immunofluorescence imaging protocol in cyclic fashion, currently permitting staining of up to ~80 unique epitopes.

Multimodal spatiotemporal phenotyping of human retinal organoid development.

Organoids generated from human pluripotent stem cells provide experimental systems to study development and disease, but quantitative measurements across different spatial scales and molecular modalities are lacking. In this study, we generated multiplexed protein maps over a retinal organoid time course and primary adult human retinal tissue. We developed a toolkit to visualize progenitor and neuron location, the spatial arrangements of extracellular and subcellular components and global patterning in each organoid and primary tissue. In addition, we generated a single-cell transcriptome and chromatin accessibility timecourse dataset and inferred a gene regulatory network underlying organoid development. We integrated genomic data with spatially segmented nuclei into a multimodal atlas to explore organoid patterning and retinal ganglion cell (RGC) spatial neighborhoods, highlighting pathways involved in RGC cell death and showing that mosaic genetic perturbations in retinal organoids provide insight into cell fate regulation.