Fig. 1. Overview of the SCAVENGE approach and applications.

a, For a given genetic trait/phenotype, the bias-corrected enrichment statistic is calculated for every single cell by integrating the PPs of fine-mapped variants and the scATAC-seq profiles. The top-ranked cells are selected as the seed cells. b, An M-kNN graph is constructed to represent cell–cell similarity, and the seed cells are projected onto this cell-to-cell graph. Network propagation scores for individual cells are defined according to the probability that the network reaches the stationary state from a number of steps of information propagation. c, Network propagation scores are further scaled and normalized to obtain the per-cell SCAVENGE TRS that represents the relevance of the trait/phenotype of interest for each single cell. Downstream analyses of functional annotation and interpretation are enabled at different levels, including for cell types, cell states and cell trajectories.