Fig. 1: ACTION decomposition robustness and interpretability.
From: A multiresolution framework to characterize single-cell state landscapes
a Performance of ACTIONet framework in recovering both identity and activity cell patterns. Cells of the same identity but with variable activity pattern form a gradient radiating from the core. b Performance of ACTIONet in identifying cell identities across 20 instances at three different noise levels (evaluating matrix H). c Performance of ACTIONet in recovering pure and interpretable identity and activity patterns or gene expression profiles (evaluating matrix W). n = 20 independent samples over each of n = 3 independent noise levels. Data are presented as mean values ±SD. d Consistency of learned (columns) and predicted (rows) patterns for the different methods tested, considering the case of 14 known patterns.
