Not getting in too deep: a practical deep learning approach to routine crystallisation image classification

Abstract

AbstractUsing a relatively small training set of ∼16 thousand images from macromolecular crystallisation experiments, we compare classification results obtained with four of the most widely-used convolutional deep-learning network architectures that can be implemented without the need for extensive computational resources. We show that the classifiers have different strengths that can be combined to provide an ensemble classifier achieving a classification accuracy comparable to that obtained by a large consortium initiative. We use eight classes to effectively rank the experimental outcomes, thereby providing detailed information that can be used with routine crystallography experiments to automatically identify crystal formation for drug discovery and pave the way for further exploration of the relationship between crystal formation and crystallisation conditions.Author summaryX-ray crystallography can provide structural information on the molecules that play important roles in cell function and allow interactions vital for drug design to be elucidated. However, the technique requires the molecules to be crystallised and obtaining sufficiently high-quality crystals can require hundreds of experiments under different conditions. Robots have been designed to perform the microscopic experiments with imaging systems to record the results over time but automated classification of all these images is becoming essential. In this work we compare different classifiers that can be used without the need for vast computational resources and show that an ensemble classifier which combines the different strengths of four deep-learning networks is able to correctly classify the experimental results into one of eight different categories 94% of the time.