STEER: decoupling kinetics with Spatial-Temporal Explainable Expert model for RNA velocity inference.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• RNA velocity analysis uses spliced and unspliced mRNA from single-cell or spatial transcriptomics to understand cell dynamics.
• Current methods struggle with heterogeneous kinetic regimes and spatial mixing, limiting accuracy and interpretability.
• Complex biological systems often exhibit mixed kinetic populations, posing challenges for existing RNA velocity tools.

Purpose of the Study:

• To develop a flexible and interpretable framework for RNA velocity inference that addresses limitations of existing methods.
• To disentangle kinetically and/or spatially mixed cell populations in complex tissues.
• To provide enhanced accuracy and interpretability for spatial transcriptomic data analysis.

Main Methods:

• Introduced STEER (Spatial-Temporal Explainable Expert model for RNA-velocity inference), integrating a spatially informed graph-attention auto-encoder.
• Employed a kinetically guided mixture-of-experts architecture to assign cells to distinct kinetic regimes.
• Inferred cell-gene-specific kinetic rates and cell-level latent time for dynamic modeling.

Main Results:

• STEER demonstrated robust performance and enhanced interpretability on synthetic and real-world datasets.
• The framework successfully disentangled kinetically and spatially mixed cell populations.
• Revealed spatiotemporally complementary immunoregulatory programs at the maternal-fetal interface in mouse uterus.

Conclusions:

• STEER offers a generalizable and explainable framework for analyzing complex spatio-temporal dynamics in tissues.
• Provides valuable insights into tissue morphogenesis, lineage specification, and tumor progression.
• Advances the field of RNA velocity inference for complex biological systems.