Integrated analysis of tissue histology with the genome-wide array and clinical data has the potential to generate hypotheses as well as be prognostic. However, due to the inherent technical and biological variations, automated analysis of whole mount tissue sections is impeded in very large datasets, such as The Cancer Genome Atlas (TCGA), where tissue sections are collected from different laboratories. We aim to characterize tumor architecture from hematoxylin and eosin (H&E) stained tissue sections, through the delineation of nuclear regions on a cell-by-cell basis. Such a representation can then be utilized to derive intrinsic morphometric subtypes across a large cohort for prediction and molecular association. Our approach has been validated on manually annotated samples, and then applied to a Glioblastoma Multiforme (GBM) cohort of 377 whole slide images from 146 patients. Further bioinformatics analysis, based on the multidimensional representation of the nuclear features and their organization, has identified (i) statistically significant morphometric sub types; (ii) whether each subtype can be predictive or not; and (iii) that the molecular correlates of predictive subtypes are consistent with the literature. The net result is the realization of the concept of pathway pathology through analysis of a large cohort of whole slide images.