GraphPCA: a fast and interpretable dimension reduction algorithm for spatial transcriptomics data

Abstract

The rapid advancement of spatial transcriptomics technologies has revolutionized our understanding on cell heterogeneity and intricate spatial structures within tissues and organs. However, the high dimensionality and noise in spatial transcriptomic data present great challenges for downstream analyses. Here, we develop GraphPCA, a novel graph-constrained, interpretable, and quasi-linear dimension-reduction algorithm tailored for spatial transcriptomic data. GraphPCA leverages the strengths of graphical regularization and Principal Component Analysis (PCA) to extract low-dimensional embeddings of spatial transcriptomes that integrate location information in nearly linear time complexity. Through comprehensive evaluations on simulated data and multi-resolution spatial transcriptomic data generated from various platforms, we demonstrate the capacity of GraphPCA to enhance downstream analysis tasks including spatial domain detection, denoising, and trajectory inference. The computational efficiency and scalability of GraphPCA facilitate the development of GraphPCA_multi, a multi-slice extension of GraphPCA that effectively captures shared tissue structures across slices. GraphPCA_multi achieved more accurate spatial domain detection than its single-slice version and other competing methods in the field. The substantial power boost enabled by GraphPCA benefits various downstream tasks of spatial transcriptomic data analyses and provides more precise insights into transcriptomic and cellular landscapes of complex tissues.