6.5 - Set of predicted biomarkers

The imaging use case addresses interoperability of large-scale image data sets, which is required for reusing and comparatively analyzing data sets generated by different sources. Cellular phenotypes are good reporters of underlying cellular processes. This observation motivates the use of experimental systems such as RNAi screens or gene knock-out mice from which the molecular underpinning of cellular phenotypes can be inferred. The same information is however difficult to get for human diseases, usually requiring costly genome-wide association studies that often end up being inconclusive. Comparing cellular phenotypes of human disease samples to those of experimental systems could be a faster and cheaper alternative by allowing for transfer of information suggesting likely molecular mechanisms or biological processes affected in the human samples. This information could then be used to predict biomarkers for the corresponding diseases. However, because annotations of cellular phenotypes are made of free text descriptions that are often ambiguous, comparison of cellular phenotypes within and across domains is currently not possible. To harmonize the annotation of cellular phenotypes, WP6 partners have developed the Cellular Microscopy Phenotype Ontology (CMPO), a species neutral ontology for describing phenotypic observations at the cellular level. CMPO was used to annotate images from gene knock-out mice, human cancer samples and RNAi screens to try and recover preselected candidate biomarkers. Analysis of the resulting annotation data uncovered different annotation practices across RNAi screens, mouse and human samples, which hamper comparisons across these imaging domains. Our observations suggest a possible need to reorganise CMPO and indicate that, in addition to annotation standards, we also need standardized annotation practices.