Alignment of Spatial Transcriptomics data with the alignProMises R package

Author(s): Daniela Corbetta,Angela Andreella,Davide Risso,Livio Finos

Affiliation(s): Department of Statistical Sciences - University of Padua



Spatial transcriptomics is a recent genome-sequencing technique that provides information on the spatial organization of tissues while simultaneously obtaining gene expression data. The application of this technology could revolutionize medical research by providing insights into the genomic basis of brain diseases. However, analysis of brains of different subjects is challenging because they are not functionally aligned. This could lead to biased results in the analysis of samples of different subjects under different biological or pathological conditions. Alignment of samples is indeed a preliminary and unavoidable step. In this project, we apply the ProMises alignment [1] to spatial transcriptomics data to demonstrate that this approach removes a major source of technical variability. Procrustes analysis is a statistical shape analysis that aligns matrices in a common reference space using similarity transformations. It has been rephrased as a statistical model in the perturbation model, which defines matrices as random rotations of a common reference matrix plus an error. However, its solution is not unique, lacking interpretability since the aligned images lose their anatomical structure. To address this issue, Andreella and Finos (2022) proposed the ProMises model, a Bayesian extension of the perturbation model that assumes a von Mises-Fisher distribution as the prior distribution for the rotation parameter. They also introduced the Efficient ProMises model, which reduces the computational load of the ProMises model in the case of high-dimensional data without loss of information and is suitable for matrices with different dimensions. Both these models provide a unique solution. Here, we implemented this approach in the R package alignProMises to apply Procrustes alignment to 10X Genomics Visium spatial transcriptomics data obtained from the dorso-lateral-prefrontal cortex of three independent neurotypical donors [2]. The package contains 3 main functions: ProMisesModel, which performs functional alignment using the ProMises model, EfficientProMises, which implements the efficient version of the ProMises transformation, and EfficientProMisesSubj, which allows for matrices with different dimensions. The results demonstrated that the ProMises transformation significantly reduced the technical variability caused by misalignment. Indeed, the analysis conducted on aligned samples revealed less false positives in differentially expressed genes, highlighting the need for alignment methods in spatial transcriptomics data analysis. ProMises alignment is a promising approach for aligning spatial transcriptomics data, which can remove technical unwanted variability and improve the accuracy of downstream analysis. Our findings demonstrate that Procrustes-alignment algorithms can be applied to spatial transcriptomics data from different donors and provide a more reliable and comprehensive view of gene expression in the brain. Further research is required to explore the potential of this approach in other settings. [1] Andreella, A., Finos, L. Procrustes Analysis for High-Dimensional Data. Psychometrika 87, 1422–1438 (2022) [2] Maynard, K.R., Collado-Torres, L., Weber, L.M. et al. Transcriptome-scale spatial gene expression in the human dorsolateral prefrontal cortex. Nat Neurosci 24, 425–436 (2021)