An information theoretic approach to detecting spatially varying genes.
An information theoretic approach to detecting spatially varying genes.
复制标题
DOI：
10.1016/j.crmeth.2023.100507
复制
发表时间：
2023-06-26
期刊：
影响因子：
--
通讯作者：
中科院分区：
文献类型：
--
作者：
A key step in spatial transcriptomics is identifying genes with spatially varying expression patterns. We adopt an information theoretic perspective to this problem by equating the degree of spatial coherence with the Jensen-Shannon divergence between pairs of nearby cells and pairs of distant cells. To avoid the notoriously difficult problem of estimating information theoretic divergences, we use modern approximation techniques to implement a computationally efficient algorithm designed to scale with in situ spatial transcriptomics technologies. In addition to being highly scalable, we show that our method, which we call maximization of spatial information (Maxspin), improves accuracy across several spatial transcriptomics platforms and a variety of simulations when compared with a variety of state-of-the-art methods. To further demonstrate the method, we generated in situ spatial transcriptomics data in a renal cell carcinoma sample using the CosMx Spatial Molecular Imager and used Maxspin to reveal novel spatial patterns of tumor cell gene expression. Spatial organization can be formalized in information theoretic terms Modern methods from deep learning enable efficient estimation of spatial information Spatial information outperforms other methods at identifying spatially varying genes Patterns of expression in renal cell carcinoma are revealed using spatial information Identifying genes with spatially coherent expression patterns is a key task in spatial transcriptomics analysis. Previously proposed methods have various shortcomings, often failing to adequately conceptualize the problem, making overly strong distributional assumptions, struggling to scale to the rapidly increasing scale of the data, or failing to demonstrate any improvement over classical spatial statistics methods. Quantifying the degree of spatial organization in a biological signal is a key building block of spatial transcriptomics analysis. Jones et al. conceptualize this in information theoretic terms, providing an interpretable and accurate tool to reveal spatial organization.
