AbstractSingle-cell resolution technologies warrant computational methods that capture cell heterogeneity while allowing efficient comparisons of populations. Here, we summarize cell populations by adding features’ dispersion and covariances to population averages, in the context of image-based profiling. We find that data fusion is critical for these metrics to improve results over the prior state-of-the-art, providing at least ~20% better performance in predicting a compound’s mechanism of action (MoA) and a gene’s pathway.
Capturing single-cell heterogeneity via data fusion improves image-based profiling