Bayesian phylogenetics with incomplete lineage tracing data

Abstract

Single-cell lineage tracing is rapidly emerging as a powerful method to track the dynamics of cell populations through time. Specifically, CRISPR-Cas9-based dynamic lineage tracing methods generate high-resolution data with the potential to investigate increasingly complex biological processes. However, single-cell readouts are inherently lossy, resulting in significant sparsity in lineage barcode alignments. Here, we present a mechanistic model of barcode loss implemented in the Bayesian framework BEAST 2, enabling phylogenetic inference from sparse datasets without the need for data filtering. Through simulations, we find that incorporating a mechanistic model of loss improves accuracy across all estimates in phylogenetic analyses by enabling previously unusable data to contribute to the inference. We further demonstrate the model's utility across two datasets: one from a monoclonal cell culture and another from murine gastruloids. Consistent with our simulation findings, applying our model to these datasets recovers higher growth rates than workflows relying on standard data filtering, confirming that such filtering can lead to an underestimation of population expansion. Crucially, we demonstrate that our framework enables inference even from datasets whose sparsity previously prevented analysis, providing a robust foundation for extracting biological insights from lineage tracing experiments.