Spatial transcriptomics (ST) technologies enable high throughput gene expression characterization within thin tissue sections. However, comparing spatial observations across sections, samples, and technologies remains challenging. To address this challenge, we develop STalign to align ST datasets in a manner that accounts for partially matched tissue sections and other local non-linear distortions using diffeomorphic metric mapping. We apply STalign to align ST datasets within and across technologies as well as to align ST datasets to a 3D common coordinate framework. We show that STalign achieves high gene expression and cell-type correspondence across matched spatial locations that is significantly improved over landmark-based affine alignments. Applying STalign to align ST datasets of the mouse brain to the 3D common coordinate framework from the Allen Brain Atlas, we highlight how STalign can be used to lift over brain region annotations and enable the interrogation of compositional heterogeneity across anatomical structures. STalign is available as an open-source Python toolkit at
Nonnegative matrix factorization (NMF) is widely used to analyze high-dimensional count data because, in contrast to real-valued alternatives such as factor analysis, it produces an interpretable parts-based representation. However, in applications such as spatial transcriptomics, NMF fails to incorporate known structure between observations. Here, we present nonnegative spatial factorization (NSF), a spatially-aware probabilistic dimension reduction model based on transformed Gaussian processes that naturally encourages sparsity and scales to tens of thousands of observations. NSF recovers ground truth factors more accurately than real-valued alternatives such as MEFISTO in simulations, and has lower out-of-sample prediction error than probabilistic NMF on three spatial transcriptomics datasets from mouse brain and liver. Since not all patterns of gene expression have spatial correlations, we also propose a hybrid extension of NSF that combines spatial and nonspatial components, enabling quantification of spatial importance for both observations and features. A TensorFlow implementation of NSF is available from
- Award ID(s):
- 2243341
- PAR ID:
- 10388861
- Publisher / Repository:
- Nature Publishing Group
- Date Published:
- Journal Name:
- Nature Methods
- Volume:
- 20
- Issue:
- 2
- ISSN:
- 1548-7091
- Page Range / eLocation ID:
- p. 229-238
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
Abstract https://github.com/JEFworks-Lab/STalign and as Supplementary Software with additional documentation and tutorials available athttps://jef.works/STalign . -
Advances in single cell transcriptomics have allowed us to study the identity of single cells. This has led to the discovery of new cell types and high resolution tissue maps of them. Technologies that measure multiple modalities of such data add more detail, but they also complicate data integration. We offer an integrated analysis of the spatial location and gene expression profiles of cells to determine their identity. We propose scHybridNMF (single-cell Hybrid Nonnegative Matrix Factorization), which performs cell type identification by combining sparse nonnegative matrix factorization (sparse NMF) with k-means clustering to cluster high-dimensional gene expression and low-dimensional location data. We show that, under multiple scenarios, including the cases where there is a small number of genes profiled and the location data is noisy, scHybridNMF outperforms sparse NMF, k-means, and an existing method that uses a hidden Markov random field to encode cell location and gene expression data for cell type identification.more » « less
-
more » « less
Abstract Single cell profiling techniques including multi-omics and spatial-omics technologies allow researchers to study cell-cell variation within a cell population. These variations extend to biological networks within cells, in particular, the gene regulatory networks (GRNs). GRNs rewire as the cells evolve, and different cells can have different governing GRNs. However, existing GRN inference methods usually infer a single GRN for a population of cells, without exploring the cell-cell variation in terms of their regulatory mechanisms. Recently, jointly profiled single cell transcriptomics and chromatin accessibility data have been used to infer GRNs. Although methods based on such multi-omics data were shown to improve over the accuracy of methods using only single cell RNA-seq (scRNA-seq) data, they do not take full advantage of the single cell resolution chromatin accessibility data.
We propose CeSpGRN (
Ce llSp ecificG eneR egulatoryN etwork inference), which infers cell-specific GRNs from scRNA-seq, single cell multi-omics, or single cell spatial-omics data. CeSpGRN uses a Gaussian weighted kernel that allows the GRN of a given cell to be learned from the sequencing profile of itself and its neighboring cells in the developmental process. The kernel is constructed from the similarity of gene expressions or spatial locations between cells. When the chromatin accessibility data is available, CeSpGRN constructs cell-specific prior networks which are used to further improve the inference accuracy.We applied CeSpGRN to various types of real-world datasets and inferred various regulation changes that were shown to be important in cell development. We also quantitatively measured the performance of CeSpGRN on simulated datasets and compared with baseline methods. The results show that CeSpGRN has a superior performance in reconstructing the GRN for each cell, as well as in detecting the regulatory interactions that differ between cells. CeSpGRN is available at
https://github.com/PeterZZQ/CeSpGRN . -
more » « less
Abstract Spatial transcripome (ST) profiling can reveal cells’ structural organizations and functional roles in tissues. However, deciphering the spatial context of gene expressions in ST data is a challenge—the high-order structure hiding in whole transcriptome space over 2D/3D spatial coordinates requires modeling and detection of interpretable high-order elements and components for further functional analysis and interpretation. This paper presents a new method GraphTucker—graph-regularized Tucker tensor decomposition for learning high-order factorization in ST data. GraphTucker is based on a nonnegative Tucker decomposition algorithm regularized by a high-order graph that captures spatial relation among spots and functional relation among genes. In the experiments on several Visium and Stereo-seq datasets, the novelty and advantage of modeling multiway multilinear relationships among the components in Tucker decomposition are demonstrated as opposed to the Canonical Polyadic Decomposition and conventional matrix factorization models by evaluation of detecting spatial components of gene modules, clustering spatial coefficients for tissue segmentation and imputing complete spatial transcriptomes. The results of visualization show strong evidence that GraphTucker detect more interpretable spatial components in the context of the spatial domains in the tissues.
Availability and implementation https://github.com/kuanglab/GraphTucker.
-
more » « less
Abstract We introduce a new method based on nonnegative matrix factorization, Neural NMF, for detecting latent hierarchical structure in data. Datasets with hierarchical structure arise in a wide variety of fields, such as document classification, image processing, and bioinformatics. Neural NMF recursively applies NMF in layers to discover overarching topics encompassing the lower-level features. We derive a backpropagation optimization scheme that allows us to frame hierarchical NMF as a neural network. We test Neural NMF on a synthetic hierarchical dataset, the 20 Newsgroups dataset, and the MyLymeData symptoms dataset. Numerical results demonstrate that Neural NMF outperforms other hierarchical NMF methods on these data sets and offers better learned hierarchical structure and interpretability of topics.
