Automated multivariate profiling of drug effects from fluorescence microscopy images

Fluorescence microscopy is a useful tool for building quantitative profiles of drug effects. Although features with rich information can be extracted from fluorescence microscopy images, most current profiling methods build profiles from the extracted features using either univariate or non-automated methods. We propose a new multivariate, automated and scalable method for building drug profiles by using a decision hyperplane. The method was evaluated by using 23 compounds belonging to four groups of known mechanisms. We produced quantitative profiles that group drugs with similar mechanisms together, and separate drugs with dissimilar mechanisms from each other. These profiles resulted in better characterizations of the drug effects than profiles obtained from a previous univariate method