This paper introduces CellTreeQM, a novel deep learning framework designed to reconstruct cell lineage trees from phenotypic features, such as gene expression profiles. The authors explain how traditional lineage tracing methods are often limited by feasibility or ethical constraints, leading to a need for alternative approaches. CellTreeQM addresses this by formulating lineage reconstruction as a tree-metric learning problem, which optimizes an embedding space for tree-graph inference. The method is benchmarked using synthetic data and lineage-resolved single-cell RNA sequencing datasets from C. elegans, demonstrating its ability to recover lineage structures with minimal supervision and limited data. The paper thoroughly discusses the challenges of phenotype-based lineage reconstruction, including the uncharacterized stochastic processes in gene expression data and the scarcity of ground-truth annotations.
References:
