Algorithmic advances in machine learning for single-cell expression analysis

Abstract

Numerous recent insights into the complex and dynamic cellular signatures of biological function, development and disease are owed to technical advances in single-cell transcriptome profiling. The deep understanding of cellular biology that single cell expression experiments can deliver comes at a cost, as the complexity and size of the data poses unique challenges to computational data analysis and algorithmic development. In this review, we highlight recent advances in machine learning that allow for efficient integration of single cell data, the identification of known and novel cell types using biological information, and the modeling of dynamic changes of cell populations over time. We summarize novel algorithmic approaches that promise robust, scalable, and explainable analysis of the cells’ multidimensional and dynamic nature. The conclusion provides a brief outlook into the potential future of machine learning–based single-cell expression analysis.