Virtual lineages: in silico lineage tracing from single-cell and spatial transcriptomes

Abstract

Experimental lineage tracing reveals histories and fates of individual cells but is costly and often impractical, particularly in primary human tissues. Gene expression differences that persist across cell divisions provide a basis for inferring ancestry from more accessible single cell transcriptomes, yet accurate lineage reconstruction remains challenging. Here, we introduce isCORAL, an in silico lineage-tracing framework that reconstructs clonal relationships and maps cell-fate dynamics from transcriptome snapshots alone. By identifying memory genes with heritable expression differences, isCORAL infers probabilistic clonal relationships and uses them to construct clone-oriented embeddings and dynamic cell-fate maps. Across diverse lineage-tracing datasets, isCORAL markedly outperforms tested alternatives in both memory-gene identification and lineage reconstruction. We use isCORAL to resolve early fate biases among transcriptionally similar hematopoietic stem cells and predict distinct developmental routes to shared outcomes. We extend isCORAL to spatial transcriptomics, validating inferred clonal relationships against an experimentally recorded lineage tree in mouse tumors and inferring early epiblast and amnion dynamics in a human embryo. Together, isCORAL enables lineage-resolved studies of development and disease from standard single cell and spatial transcriptomes, and provides an explicit representation of lineage memory for future virtual cell models.