Deep Lineage: Single-cell lineage tracing and fate inference using deep learning

Recent advances in single-cell RNA sequencing and lineage tracing techniques have provided valuable insights into the temporal changes in gene expression during development, tumor progression, and disease onset. However, there are few computational methods available to analyze this information to help understand multicellular dynamics. We introduce Deep Lineage, a deep-learning method for analyzing time-series single-cell RNA sequencing with matched lineage tracing data. Our method accurately predicts early cell fate biases and gene expression profiles at different time points within a clone, surpassing existing methods in fate prediction accuracy. We further validate our model using LineageSim, a fate-aware lineage simulator, demonstrating robustness to technical noise such as barcode dropout and clone misidentification. Additionally, we apply Deep Lineage to in silico perturbations in cellular reprogramming and hematopoiesis to identify key genes and pathways associated with cell fate determination and propose testable perturbation hypotheses for experimental follow-up.