tricycle

Tricycle infers and visualizes cell cycle position from single-cell RNA-seq data by projecting new data into a pre-learned, biologically interpretable cell cycle embedding using transfer learning.

Key Features:

• Transfer learning projection: Projects new single-cell RNA-seq data into a previously learned cell cycle space rather than learning a dataset-dependent embedding.
• Pre-learned cell cycle space: Provides a reference embedding trained on fixed reference data that enables inference of cell cycle time in human and mouse single cell samples.
• Cell cycle time visualization: Includes functions to visualize inferred cell cycle time on different embeddings.
• Custom reference construction: Offers functions to build new cell cycle references from user-provided data.
• High-resolution position prediction: Predicts cell-specific position in the cell cycle based on data projection into the reference embedding.
• Cross-context generalization: Predictions generalize across multiple cell types, tissues, species, and sequencing assays using internal controls available for any dataset.
• Scalability: Applicable to atlas-level single-cell RNA-seq data.

Scientific Applications:

• Cell cycle position inference: Estimates high-resolution cell-cycle position from single-cell RNA-seq data without requiring gold-standard experimental assays.
• Confounding factor control: Enables identification and accounting for cell cycle as a confounding variable in single-cell studies.
• Cross-species analysis: Supports cell cycle inference in both human and mouse single-cell samples.
• Large-scale atlas analysis: Facilitates cell cycle analysis in large-scale single-cell RNA-seq atlases across diverse biological contexts.

Methodology:

Tricycle estimates a cell-cycle embedding using principal component analysis of periodic functions on a fixed reference dataset, projects new data into this reference embedding, and predicts cell-specific position in the cell cycle based on the data projection.