Automation and High-Throughput Technologies
Advanced imaging technologies to bridge the gap between high-content and high-throughput
The primary reasons for drug failure in the clinic are a lack of efficacy, and safety. Therefore, in order to drive a better understanding of disease biology and improve translation, cellular imaging assays in early discovery need to be increasingly complex, utilising multiple biomarkers to label several proteins in a pathway, and to quantify multiple sub-populations.
The field of image analysis has been transformed by the explosion of machine learning and AI methods, and we are now leveraging recent developments to maximise the information we get from imaging data and enable new experimental approaches. A key limitation of machine learning, and particularly deep learning models, is the requirement for large amounts of annotated training data. We have developed an active learning framework for efficient training data generation, alongside unsupervised phenotype discovery approaches, to build models which can quantify the full complexity of cellular screening data.
We are also integrating label-free phase contrast imaging into our cellular screens. A large amount of information on cellular morphology is contained in the phase contrast images, which do not take up a fluorescent colour channel, but human interpretation is very difficult. By training a deep neural network to find and segment nuclei and cells from phase contrast images alone, nuclear and cell markers are no longer required. This allows multiple biomarkers to be combined into a single screen, enabling more complex biology for less cost.