Pre-phaser: Precise cell-cycle phase detector for scRNA-seq

Precise cell cycle phase identification in scRNA-seq improves differential expression analysis. However, the sparseness of scRNA-seq data contributes to the challenge of combining external cell cycle information with gene expression for precise cell phase identification. Existing techniques select their own set of marker genes for each coarse cell cycle phase, and assign cells with enriched average marker expression to this coarse phase. We observe that precise points along the cell cycle are associated with time course points from microarray experiments that identify cycling genes. In this work, we present the effectiveness of using k-nearest neighbors (kNN) to predict coarse cell cycle phase, and extend the kNN methodology to present the first method to identify precise cell cycle phases by correlating single cell transcript counts for significant cycling genes with time course points. We demonstrate that tuned kNN outperforms existing methods in assigning coarse cell cycle phase, getting an average F1 score of 0.641 on four previously published scRNA-seq datasets. We then describe Pre-Phaser, which establishes a general computational approach for precise cell phase assignment using kNN. Our k-fold cross-validation has an accuracy of 0.872 in precise prediction, and conversion to coarse phase yields an F1 score of 0.716. Our results motivate research into precise precise cell phase assignment to enable fine-tuned differential expression analyses.