A Probabilistic Model for Cell Population Phenotyping Using HCS Data
Density plot of cell population descriptors averaged over wells (panel (a) to (e)) and log likelihood (panel (f)) given by the model trained on negative controls. Positive controls are very different from negative controls. It is easy to distinguish them from negative controls only looking at cell count. The log likelihood given by the model separates the two type of controls. We observe that the discriminative power of the univariate descriptors is not lost when considering the model likelihood.