STgcor: A Distribution-Based Correlation Measurement Method for Spatial Transcriptome Data

Abstract

AbstractSpatial transcriptome technology can provide the transcript profiles of different regions in a tissue sample while preserving the positional information of spots. Based on spatial transcriptome data, the construction of gene regulatory networks can help researchers to identify gene modules and understand the biological process. The correlation measurement is the basis of the construction of gene regulatory networks. Typical correlation coefficients, such as the Pearson correlation coefficient and Spearman correlation coefficient, are hard to be applied to spatial transcriptome data, due to the outliers and sparsity in spatial transcriptome data. In this study, by observing the distribution of gene-pair expression values, we propose a novel gene correlation measurement method for spatial transcriptome data, named STgcor. In STgcor, a gene pair (X, Y) expressed on spots is represented as a vertex in a two-dimensional plane consisting of the gene pair vectors X and Y as coordinate axes. We calculate the joint probability density of Gaussian distributions of the gene pair (X, Y) to exclude outliers. Then, to overcome the sparsity of spatial transcriptome data, the correlation between the gene pair (X, Y) is measured based on the degree, trend, and location of aggregation of the distribution of gene-pair expression values. To validate the effectiveness of the STgcor, STgcor and two other correlation coefficients were applied to a weighted co-expression network analysis method on two spatial transcriptome datasets published by 10x genomics including the breast cancer dataset and prostate cancer dataset. They were evaluated based on the gene modules identified from the corresponding gene co-expression networks. The results showed that STgcor combined with the weighted gene co-expression network analysis method can discover special gene modules and some pathways related to cancer.KeywordsSpatial transcriptome dataCorrelation measurementGene module identificationGene co-expression networkDistribution of gene-pair expressionProbability density