Profiling thousands of single cell transcriptomes is routine, yet cell prioritization based on response to biological perturbations is challenging and confounded by clustering, normalization and dimensionality reduction strategies. We developed a scoring approach independent of these obstacles that unbiasedly identifies distinct transcriptomes within a set based on missing data patterns, allowing cell prioritization and feature selection for downstream analysis. Our method applied to reveals a metabolic shift that marks the transition between the amoeboid and aggregated states of this model organism.