Knowledge-driven analysis of image-based genetic screens using deep learning (360G-Wellcome-204724_Z_16_Z)

Characterising gene functions is important for understanding life at the molecular level, and has a great impact on pharmacological and biomedical studies. Genetic screens that utilise High Throughput Imaging (HTI) have proved to be a powerful tool for studying gene functions by monitoring phenotypic changes in genetically modified cells. Challenges in analysing HTI datasets have significantly hindered knowledge discovery from such rich datasets. As HTI datasets are now acquired on a routine basis, there is a great need for generalisable analysis methods. Deep learning has revolutionised computer vision as it can automatically extract features and classify raw images without the need for any image preprocessing. I will develop deep learning methods to automatically discover cellular phenotypes and infer gene functions based on phenotypic similarity. I will build an integrative framework that utilises Functional Annotations (FAs) from various biological databases to annotate and classify images. I will apply these methods to five genome-wide datasets to predict tissue-specific gene functions based on the phenotypes of various cellular structures. The outcome of this work will be robust and generalisable HTI analysis methods that associate phenotypes to gene functions using deep learning as well as the discovery of novel gene functions and associated phenotypes.