Integrative spatial analysis of cell morphologies and transcriptional states with MUSE

Abstract

Spatial transcriptomics enables the simultaneous measurement of morphological features and transcriptional profiles of the same cells or regions in tissues.Here we present multi-modal structured embedding (MUSE), an approach to characterize cells and tissue regions by integrating morphological and spatially resolved transcriptional data. We demonstrate that MUSE can discover tissue subpopulations missed by either modality as well as compensate for modality-specific noise. We apply MUSE to diverse datasets containing spatial transcriptomics (seqFISH+, STARmap or Visium) and imaging (hematoxylin and eosin or fluorescence microscopy) modalities. MUSE identified biologically meaningful tissue subpopulations and stereotyped spatial patterning in healthy brain cortex and intestinal tissues. In diseased tissues, MUSE revealed gene biomarkers for proximity to tumor region and heterogeneity of amyloid precursor protein processing across Alzheimer brain regions. MUSE enables the integration of multi-modal data to provide insights into the states, functions and organization of cells in complex biological tissues.

Multi-task learning for single-cell multi-modality biology

“…In applications, noises arising from different sources, such as the sequencing dropout effect and feature measurement error in multi-modality biological datasets, typically affect the performance of the network. Therefore, we tested the performance of UnitedNet in simulated biological datasets with different controlled noise levels 27 and simulated multi-modality imaging datasets 31,33 . The results show that the capability of UnitedNet to identify and suppress noise enables it to outperform the state-of-the-art models (Extended Fig.…”

“…MUSE simulated dataset. We apply the simulator in MUSE 27 to simulate two-modalities input to assess the robustness of UnitedNet with one low-quality. We simulate 11 two-modality datasets with 1,000 cells and 10 cell types.…”

“…We have applied this network to various multi-modality datasets (Fig. 1a), including (i) simulated multi-modality data with ground truth labels 26,27 ; (ii) simultaneously measured transcriptomics and intracellular electrophysiology 4,28 (multi-sensing data), (iii) simultaneously profiled transcriptomics and DNA accessibility 5,7 (multi-omics data), and (iv) spatially resolved transcriptomics and proteomics 29,30 (spatial-omics data). The results show superior performances in both tasks, achieving better unsupervised and supervised joint group identification and cross-modal prediction compared with other state-of-the-art methods.…”

Improving and evaluating deep learning models of cellular organization

“…Fluorescence microscopy is a fundamental and powerful approach in the life sciences that is used to observe cells and their structures by labeling them with fluorophores or fluorophore-conjugated antibodies [Lichtman and Conchello, 2005]. Deep learning applications to fluorescence microscope images have included reconstruction of super-resolution images [Ouyang et al ., 2018], cell segmentation [Stringer et al ., 2021, Greenwald et al ., 2021], integrative tissue analysis [Maric et al ., 2021, Bao et al ., 2022] and various augmented microscopy approaches [Wang et al ., 2021].…”

VistoSeg: processing utilities for high-resolution Visium/Visium-IF images for spatial transcriptomics data

“…However, these images represent a rich, untapped resource for downstream analyses. For example, nuclei segmentation of H&E images can be used to estimate the number of nuclei in each Visium spot or to identify cell types based on classic morphologies (4,5). Analysis of histological images can also identify spots containing a single cell, or spots enriched in specific cellular compartments (i.e.…”