Universal prediction of cell-cycle position using transfer learning.

Area of Science:

• Single-cell genomics
• Computational biology
• Cell cycle regulation

Background:

• The cell cycle is crucial for cell replication and division.
• Single-cell RNA sequencing (scRNA-seq) offers high-resolution measurements of cellular processes.
• A universal method for high-resolution cell cycle inference from scRNA-seq data is lacking.

Purpose of the Study:

• To develop a robust and universally applicable computational tool for inferring cell cycle position from scRNA-seq data.
• To address limitations of existing methods for cell cycle inference.

Main Methods:

• Development of 'tricycle', an R/Bioconductor package.
• Leveraging cell cycle biology, principal component analysis of periodic functions, and transfer learning.
• Estimating a cell-cycle embedding using a fixed reference dataset and projecting new data.

Main Results:

• Tricycle accurately predicts cell-specific positions in the cell cycle.
• Performance of tricycle compares favorably to gold-standard experimental assays.
• Tricycle predictions generalize across diverse datasets, including multiple cell types, tissues, species, and sequencing assays.

Conclusions:

• Tricycle provides a scalable and generalizable method for cell cycle inference from scRNA-seq data.
• Applicable to large-scale datasets, including single-cell atlases.