TRIO-AI: Hybrid temporal graph, ODE, and VAE modeling for high-resolution cellular trajectory inference in liver injury

Resolving dynamic cellular transitions at single-cell resolution is essential for understanding complex biological processes in development, disease, and regeneration. However, existing trajectory inference methods struggle to capture heterogeneous temporal dynamics, particularly for rare or transitional cell populations critical to injury and repair responses. Here, we present TRIO-AI, a hybrid computational framework that synergistically integrates three complementary approaches: Temporal Graph Neural Networks (Temporal GNN) for branching detection, Neural Ordinary Differential Equations (Neural ODEs) for continuous state flow modeling, and Time-Variational Autoencoders (Time-VAE) for density-based transitional state identification. We applied TRIO-AI to single-cell RNA sequencing data from mouse liver ischemia-reperfusion (IR) injury across multiple timepoints (days 0, 1, 3, and 5 post-injury), comprehensively characterizing major hepatic macrophage populations. TRIO-AI identified a distinct transitional macrophage population (MPs_3) that peaks at 48 hours post-injury and exhibits coordinated signatures for lipid handling (LRP1, LRP6, ABCA1, LDLR, SCARB1), efferocytosis (MARCO, MERTK, MSR1), and extracellular matrix engagement (ITGA9, ITGB1, SDC2). This population serves as a critical integrator of multicellular pro-reparative signals through APOE-LRP1 and FN1-ITGA9 signaling axes. Spatial transcriptomics validation confirmed MPs_3 localization to peri-necrotic borders with dramatic enrichment of APOE-LRP1 co-localization at 48 hours (19 pairs) compared to 24 hours (3 pairs). Comparative analysis against five state-of-the-art methods demonstrated TRIO-AI’s superior performance in detecting transitional states, accurately reconstructing branching trajectories, and identifying off-path populations missed by conventional approaches. This framework provides a powerful platform for dissecting cellular dynamics in complex biological systems and reveals new insights into macrophage-mediated tissue repair mechanisms.