Should we build single-cell lineage trees from gene expression data?

Gene expression data has been proposed as a natural single-cell lineage marker. Here, we critically examine the feasibility of reconstructing lineage trees from single-cell transcriptomic data using both modelling and empirical data. We first introduce a notion of neutrality for transcriptomic data, and then, under a model for neutral gene expression, establish theoretical bounds for accurate lineage tree reconstruction. Our findings indicate that reconstruction guarantees for even small trees or sub-trees require thousands of independent, neutral traits - a condition that is likely rarely met in practice due to the dominance of non-neutral developmental signals. Furthermore, errors introduced by measurement sampling have the potential to destroy any existing lineage signal. We conclude that gene expression data has limited potential as a natural lineage recorder and should not be used for phylogenetic lineage tree inference without further, rigorous validation.