GlaxoSmithKline reported that of the experiments performed in pharmaceutical research, up to 70% result in an image as the experimental result. These images are generated in nearly every phase of pharmaceutical research including target identification, genomics, proteomics, cell biology, pathology, autoradiography and more.
Although an interpretation of the experimental image result is made at the time of the experiment, mining the image data to do comparative assessments has proven problematic.
This presentation will discuss the architecture and practice of managing image data to enable mining. Mining is enabled from image content (visual features of the image) and image context (metadata about the image.) A web based system built on Oracle was developed to enable access and operation of the system from any java enabled browser. This provides pharmaceutical companies the ability to share data within departments, within project teams, across sites and with collaborators to review experimental results. Further, image data can be related to other experiment data to uncover relationship in say, toxicity, as a function of the properties of candidate drug compounds.
An application to extract visual features from an image to deduce protein expression relative to a control will also be discussed. Difficulty of the algorithmic approach to this problem will be reviewed along with a method to resolve historical problems with spot detection and alignment of multiple 2D electrophoresis gel images.
