scTour

scTour infers developmental pseudotime, delineates vector fields in transcriptomic latent space, and predicts cellular dynamics from single-cell genomics data using a deep learning architecture that minimizes batch effects.

Key Features:

• Batch-Insensitive Analysis: Minimizes the influence of batch effects across single-cell genomic datasets to support consistent inference and prediction.
• Developmental Pseudotime Estimation: Estimates developmental pseudotime to map cells along trajectories representing progression through developmental stages.
• Vector Field Delineation: Delineates vector fields within the transcriptomic latent space to capture directional changes in gene expression during state transitions.
• Transcriptomic Latent Space Mapping: Maps high-dimensional transcriptomic profiles into a lower-dimensional latent space to reveal underlying biological structure.
• Prediction of Cellular Dynamics: Predicts dynamics of unseen cellular states and independent datasets to model future cell behavior.
• Deep Learning Architecture: Uses a unified deep learning framework that jointly performs pseudotime estimation, vector field delineation, and latent-space mapping.

Scientific Applications:

• Developmental Biology: Characterizing cellular differentiation and maturation trajectories through pseudotime and vector-field analyses.
• Disease Modeling: Modeling disease-associated cell-state transitions and altered cellular dynamics.
• Regenerative Medicine: Informing regenerative strategies by predicting future cell states and developmental outcomes.
• Cross-dataset Prediction: Predicting cellular dynamics across independent datasets and experimental batches (demonstrated across 19 datasets).

Methodology:

Simultaneously estimates developmental pseudotime, delineates vector fields, and maps the transcriptomic latent space within an integrated deep learning framework.