Computational assignment of cell-cycle stage from single-cell transcriptome data.

Area of Science:

• Computational Biology
• Genomics
• Cell Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides insights into cellular states and heterogeneity.
• Accurate computational methods are needed to analyze scRNA-seq data, particularly for cell-cycle stage identification in dividing cell populations.

Purpose of the Study:

• To compare the performance of established supervised machine learning methods and a custom predictor for cell-cycle stage allocation using transcriptomic data.
• To assess the impact of normalization strategies and prior knowledge on the accuracy of cell-cycle classification.

Main Methods:

• Evaluation of five supervised machine learning algorithms and one custom predictor.
• Testing methods on published scRNA-seq datasets.
• Analysis of different data normalization strategies and the incorporation of prior knowledge (cell-cycle annotated genes).

Main Results:

• A PCA-based approach and the custom predictor demonstrated superior performance in cell-cycle stage allocation.
• Classification accuracy was significantly influenced by normalization techniques and the use of prior knowledge.
• Robust capture of transcriptional cell-cycle signatures across diverse cell types, organisms, and experimental protocols requires prior knowledge and rank-based normalization.

Conclusions:

• Effective cell-cycle stage determination from scRNA-seq data relies heavily on appropriate normalization and the integration of prior biological knowledge.
• The developed PCA-based and custom methods offer robust solutions for cell-cycle classification in single-cell transcriptomic studies.