Inferring Associations with Cell Morphology
The Laplacian Score is a statistical test implemented in CAJAL to determine whether differences in a numerical feature assigned to cells, \(f : G\to \mathbb{R}\), such as the expression of a gene or the genotype of the cell in a given locus, are related to differences in cell morphology. Specifically, the Laplacian Score answers the question: if \(x\) and \(y\) are two cells with similar morphology, are \(f(x)\) and \(f(y)\) closer on average than if \(x\) and \(y\) were chosen randomly?
To perform this analysis, CAJAL uses the Gromov-Wasserstein distance between every pair of cells to construct an undirected graph \(G\) where nodes represent cells and edges connect cells with distances less than \(\varepsilon\), a user-specified positive real parameter. The Laplacian score of \(f\) with respect to the graph \(G\) is positive number defined by
where \(E(G)\) is the set of edges in the graph, \(i,j\) range over nodes of \(G\), and \(\operatorname{Var}_G(f)\) is the weighted variance of f, where the weight of node \(i\) is proportional to the number of neighbors of \(i\) in \(G\). When the Laplacian Score is close to zero, this indicates that the values of \(f\) tend to be similar between connected cells.
To test the significance of the Laplacian Score, CAJAL provides a permutation test that shuffles the values of \(f\) across the nodes of \(G\) to generate a null distribution, from which a p-value can be computed. Additionally, CAJAL supports regression analysis to account for the influence of other covariates, \(g_1,\dots,g_n\), defined on \(G\). Users can fit a multivariate linear regression model to remove the dependence of \(C_G(f)\) on \(C_G(g_1),\dots, C_G(g_n)\), and evaluate whether the Laplacian Score of \(f\) is below what would be expected from the covariate features.
Overall, the Laplacian Score implemented in CAJAL provides a flexible approach for analyzing the relationship between cell morphology and numerical features, with the ability to account for other covariates and assess statistical significance.
More information about the theoretical foundations of the Laplacian score can be found at:
- Govek, K. W., et al. CAJAL enables analysis and integration of single-cell morphological data using metric geometry. Nature Communications 14 (2023) 3672.
- Govek, K. W., Yamajala, V. S., and Camara, P. G. Clustering-Independent Analysis of Genomic Data using Spectral Simplicial Theory. PLOS Computational Biology 15 (2019) 11.
- He, X., Cai, D., and Niyogi, P. Laplacian Score for Feature Selection In Advances in neural information processing systems (2005) 507-514.