SLAS 2019 AI in Process Automation Symposium

AI and deep learning through neuronal networks have become a staple in today’s data analysis. Especially in image analysis, neuronal networks have become very useful with facial recognition being a prominent example. AI has several strong points over conventional image analysis methods – such as the ability to mimick ability of the human brain/vision allowing for accurate flagging of minor changes that are otherwise hard to isolate through conventional image analysis. The implementation of deep learning from scratch through existing platforms such as Kerus and Tensorflow offers a relatively simple route towards testing the potential of AI in the context of high content screening (HCS). In this talk we will review the general outline of an implementation of Kerus/Tensorflow on a local workstation using Python with a GPU platform to lend computational power. We will further review two projects where AI has been pivotal in insuring success. The first project deals with setting up a platform in C. Elegans with a HCS readout genotoxicity that was easily solvable through conventional HCS analysis like Metamorph because of the lack of ability to capture subtle features. The second project deals with a HCS screen for modulators of host factors that make the cell resistant against a bacterial toxin. Here the subtlety of the phenotype was also difficult to capture using standard image analysis. Importantly, AI identified compounds that were active against the toxin and enabled us to directly translate our findings directly into a pre-clinical model of the disease.