Objective

In drug discovery a constant increase of phenotypic high-content screens is observed. Establishing them as a base technology has been hampered by the complexity of image analysis, which requires costly experts and their involvement for many weeks per assay. Here, we report on an innovative Deep Learning approach for automating image analysis. This approach will enable life science companies to broadly implement phenotypic screens and speed/scale-up the related projects. We introduce an analysis workflow using deep learning which minimizes the need for expertise and time, both in assay development and later at production stage. Where time is limited, it can provide more differentiated endpoints than traditional image analysis, especially in MOA studies or assay development. This standardized workflow encompasses visualizations to define known phenotypes and detect new ones; methods for efficient training data generation and curation; training of deep learning networks for subsequent classification of high-content screening image sets in production runs. We illustrate its application in a pharma case study, showing good quality of results on all levels (e.g. classification accuracy and potency). We also compare the time and expertise investment between this workflow and a traditional approach. This Deep Learning technology enables speedy, scaled-up analysis and delivers on the complexity typical for phenotypic assays in today’s research.