by Till Richter, Mojtaba Bahrami, Yufan Xia, David S. Fischer, Fabian J. Theis • 1 month ago
Self-supervised learning (SSL) has proven effective in single-cell genomics (SCG) for extracting insights from large, unlabelled datasets. This study benchmarks various SSL methods, particularly masked autoencoders, against traditional techniques, revealing SSL's strengths in transfer learning and cross-modality prediction. The research identifies optimal scenarios for SSL application, emphasizing its utility in accurately predicting cell types and integrating diverse datasets, thus enhancing biological analysis and understanding of complex data structures in SCG.