Fig. 2.

Comparisons between traditional clustering based analysis and topic model  based analysis for single-cell CRISPR screening data. a Difference between traditional clustering based analysis and topic model-based analysis for single-cell CRISPR screening data when a perturbation has a significant phenotype on the cells. Both analyses can detect such phenotype change (see the cell sample with red dotted line). b Difference between traditional clustering-based analysis and topic model-based analysis for single-cell CRISPR screening data when a perturbation has a subtle phenotype on the cells. Topic modeling calculates a topic probability profile for each sample while traditional clustering just makes a hard assignment of the sample to each cluster. Therefore, in this way, topic-model-based analysis can detect such phenotype change based on the change of topic probability profile with and without perturbation, while traditional clustering based analysis failed to detect such subtle phenotype change (see the cell sample with red dotted line)