tricycle

tricycle predicts continuous cell cycle positions from single-cell RNA sequencing (scRNA-seq) data by projecting samples into a transfer-learning reference embedding.

Key Features:

• Transfer Learning: Employs transfer learning with a fixed reference dataset to create a dataset-independent cell-cycle embedding and to project new scRNA-seq data into that space.
• Mathematical Basis: Uses principal component analysis (PCA) tailored for periodic functions to represent cell cycle dynamics and assign cyclic positions.
• Scalability and Generalization: Scales to large scRNA-seq datasets and generalizes across cell types, tissues, species, and sequencing assays.
• Accuracy and Validation: Produces high-resolution predictions with accuracy comparable to gold-standard experimental assays and validates predictions using internal controls within datasets.
• Implementation: Distributed as an R package.

Scientific Applications:

• Cell Cycle Analysis: Resolving continuous cell cycle progression at single-cell resolution.
• Comparative Studies: Enabling comparison of cell cycle states across different tissues, species, and experimental conditions.
• Integration with Other Data Types: Incorporating inferred cell cycle positions into scRNA-seq analyses and identifying cell-cycle-related confounders.

Methodology:

Uses a fixed reference dataset and transfer learning to project new scRNA-seq samples into a dataset-independent cell-cycle embedding, applies PCA adapted for periodic functions to assign continuous cell cycle positions, and validates predictions using internal dataset controls; implemented as an R package.