AI detection of druggable features from high content imaging data

The value of imaging data to drug discovery and disease research is now being realised. A wave of several large repositories storing high and low density bio-image data are now publicly available for use. Two such big-data resources that are of great interest to the biopharmaceutical industry are the Cancer Imaging Consortium, Broad Institute JUMP Consortium cellular perturbations archive which contains more than 1 billion cells with 140,000 perturbations and the DeepCell normal tissues imaging atlas. These resources can be data-mined and modelled to predict the effect of chemical and genetic perturbations on cells.

In morphological profiling, quantitative data are extracted from microscopy images of cells to identify biologically relevant similarities and differences. Rich feature data (including measures of size, shape, texture and intensity) produce profiles suitable for the detection of subtle phenotypes and cell lineage information.

The project will involve extracting valuable biomarker features from morphological images in normal and perturbed cellular states. William will investigate the correlation of these features with other 'omics modalities and phenotypes, and will develop expertise in the field of deep learning to interpret the effects of perturbations on cells.