Knowledge-driven analysis of image-based genetic screens using deep learning

An important question in molecular biology is what are the functions of different genes and how they determine the organism phenotype in health and disease. Genetic screens allow analysis of  gene functions by observing the phenotype of cells when genes are perturbed. Cells can be imaged using automated microscopy after the perturbation of thousands of genes. However, the analysis of the resulting imaging datasets toward inferring gene function remains challenging, labour intensive and highly sensitive to experimental settings.

I propose to develop methods that transform the analysis of high throughput imaging data from ad-hoc pipelines to a systematic and generalisable framework. I will build on the advances in computer vision to develop methods that can discover phenotypic effects of genetic perturbations with minimal human intervention. These methods will be coupled with a bioinformatics platform to automatically link phenotypes to gene functions. As the methods learn directly from raw images they can be applied to various bioimaging studies. Developing such standardised methods is crucial for enabling comprehensive analysis of imaging datasets towards systematic inference of gene functions at different system levels.

This research will advance our knowledge about gene functions and has the potential for identifying new candidate therapeutic target genes.