Figure 1

Workflow of the method. Each knockdown movie comes along with at least three negative control movies. A: Cells in Time Lapse Movies are segmented and identified. B: Tracking of the cells using CellProfiler results in trajectories of cells. C: Morphological features are calculated for each cell, which gives a sequence of high dimensional feature vectors for each trajectory. D: An HMM with Gaussian emission densities is learned on the pooled feature vectors of each knockdown - negative control pair (for clarity of presentation, the HMM has only 3 instead of 6 states actually used in the analysis). E: Cell morphologies are clustered to phenotype classes by the HMM Viterbi path assignment. Knockdown-specific phenotypes are extracted by comparing the phenotype class frequencies. F: Knockdown-specific phenotypes of many RNAi experiments are pooled and clustered by a Gaussian Mixture model. The cluster proportions of a knockdown define its phenotypic fingerprint.