Abstract

Imaging assays that drive early discovery of targets, drugs and mechanisms have grown in complexity, requiring more sophisticated yet easy-to-use and enterprise-scale analysis solutions. Artificial intelligence-based image analysis methods have the potential to deliver high quality results in an automated, unbiased way and yield biological insights not accessible by traditional image analysis techniques. AstraZeneca deployed Genedata Imagence, a new enterprise-scale software solution for image analysis based on deep learning. Deep learning-based approaches require labelled training data to train networks. Imagence made it easy to generate large training sets for neural network training. No image analysis or machine learning expertise was required, and trained networks can be easily used to classify cellular phenotypes at unlimited scale. Imagence was deployed onto a hybrid AWS cloud/on-premise platform and supports users of multiple instruments across different sites – a key attainment, given the data-intensive nature of deep learning workflows. In the first year following implementation, AstraZeneca has run a series of early-adopter projects spanning different disease areas and observed four primary benefits: 1) Enabling assays and endpoints that were previously challenging or not feasible; 2) Increased assay robustness; 3) Identification of novel phenotypes of value; 4) Time saving through both facilitated assay development and automated analysis of production-level screens. These scientific and efficiency gains result in accelerated drug discovery with increased mechanistic insight.