DestinyNet: A deep-learning framework for cell-fate analysis from lineage-tracing single-cell RNA sequencing data

<h2>Summary</h2> Unraveling cell-development dynamics, including lineage commitment, differentiation, and disease progression, is fundamental to biology. Despite advances in single-cell omics and barcoding technologies, comprehensive frameworks for accurate, robust, and scalable cell-fate analysis using lineage-tracing single-cell RNA sequencing (LT-scSeq) data remain limited. We introduce DestinyNet, a multi-task deep-learning framework addressing three key challenges: (1) fate clustering, integrating fate and cell-type information; (2) fate flow, depicting dynamic pseudotime trajectories with fate information; and (3) fate prediction, identifying early-stage cell-fate biases. DestinyNet enables end-to-end cell representation learning through cell-relation triplets and is robust across various LT-scSeq data types, including static, cumulative, and dynamic barcoding with single or multiple time points. Experiments on diverse datasets, including hematopoiesis differentiation and fibroblast reprogramming (<i>in vitro</i> and <i>in vivo</i>), demonstrate DestinyNet's effectiveness in multiple fate-analysis tasks.