Here we introduce repeated decision stumping, to distill simple models from single cell data. We develop decision trees of depth one – hence ‘stumps’ – to identify in an inductive manner, gene products involved in driving cell fate transitions, and in applications to published data we are able to discover the key-players involved in these processes in an unbiased manner without prior knowledge. The approach is computationally efficient, has remarkable predictive power, and yields robust and statistically stable predictors: the same set of candidates is generated by applying the algorithm to different subsamples of the data.

Single cell transcriptomics has laid bare the heterogeneity of apparently identical cells at the level of gene expression. For many cell-types we now know that there is variability in the abundance of many transcripts, and that average transcript abun-dance or average gene expression can be a unhelpful concept. A range of clustering and other classification methods have been proposed which use the signal in single cell data to classify, that is assign cell types, to cells based on their transcriptomic states. In many cases, however, we would like to have not just a classifier, but also a set of interpretable rules by which this classification occurs. Here we develop and demonstrate the interpretive power of one such approach, which sets out to establish a biologically interpretable classification scheme. In particular we are interested in capturing the chain of regulatory events that drive cell-fate decision making across a lineage tree or lineage sequence. We find that suitably defined decision trees can help to resolve gene regulatory programs involved in shaping lineage trees. Our approach combines predictive power with interpretabilty and can extract logical rules from single cell data.