ProjMC$^2$: Scalable and Stable Posterior Inference for Bayesian Spatial Factor Models with Application to Spatial Transcriptomics (2506.01098v1)

Abstract: Factor models exhibit a fundamental tradeoff among flexibility, identifiability, and computational efficiency. Bayesian spatial factor models, in particular, face pronounced identifiability concerns and scaling difficulties. To mitigate these issues and enhance posterior inference reliability, this work proposes Projected Markov Chain Monte Carlo (ProjMC$^2$), a novel Markov Chain Monte Carlo (MCMC) sampling algorithm employing projection techniques and conditional conjugacy. ProjMC$^2$ is showcased within the context of spatial factor analysis, significantly improving posterior stability and MCMC mixing efficiency by projecting posterior sampling of latent factors onto a subspace of a scaled Stiefel manifold. Theoretical results establish convergence to the stationary distribution irrespective of initial values. Integrating this approach with scalable univariate spatial modeling strategies yields a stable, efficient, and flexible modeling and sampling methodology for large-scale spatial factor models. Simulation studies demonstrate the effectiveness and practical advantages of the proposed methods. The practical utility of the methodology is further illustrated through an analysis of spatial transcriptomic data obtained from human kidney tissues, showcasing its potential for enhancing the interpretability and robustness of spatial transcriptomics analyses.