Inferring causal trajectories from spatial transcriptomics using CASCAT

Spatial trajectory inference models cell differentiation and state dynamics within tissues by integrating spatial information. Existing spatial trajectory inference methods depend on similarity-based cell graphs constructed from spatial proximity, with less attention to the Markovian property in cell state transitions. In this study, we introduce CASCAT, a tree-shaped structural causal model with the Markovian property integrated to infer a unique cell differentiation trajectory, addressing challenges posed by Markov equivalence in high-dimensional and nonlinear data. CASCAT outperforms six state-of-the-art single-cell RNA sequencing (scRNA-seq)-oriented methods across 10 simulated and seven real scRNA-seq datasets and exceeds three leading spatial trajectory inference methods on 13 real spatial transcriptomics datasets from multiple platforms. In the mouse inner olfactory bulb, CASCAT accurately differentiates maturation trajectories among specific cell types and reveals the Wnt signaling pathway by removing conditionally independent connections. Furthermore, by modeling post-treatment cancer cell trajectories through in silico simulations, CASCAT predicts drug responses in oral squamous cell carcinoma with a 6.8% increase in precision compared to RNA velocity-based methods, contributing to advances in computer-assisted drug discovery.