Using OWL reasoning to support the generation of novel gene sets for enrichment analysis.

BACKGROUND: The Gene Ontology (GO) consists of over 40,000 terms for biological processes, cell components and gene product activities linked into a graph structure by over 90,000 relationships. It has been used to annotate the functions and cellular locations of several million gene products. The graph structure is used by a variety of tools to group annotated genes into sets whose products share function or location. These gene sets are widely used to interpret the results of genomics experiments by assessing which sets are significantly over- or under-represented in results lists. F Hoffmann-La Roche Ltd. has developed a bespoke, manually maintained controlled vocabulary (RCV) for use in over-representation analysis. Many terms in this vocabulary group GO terms in novel ways that cannot easily be derived using the graph structure of the GO. For example, some RCV terms group GO terms by the cell, chemical or tissue type they refer to. Recent improvements in the content and formal structure of the GO make it possible to use logical queries in Web Ontology Language (OWL) to automatically map these cross-cutting classifications to sets of GO terms. We used this approach to automate mapping between RCV and GO, largely replacing the increasingly unsustainable manual mapping process. We then tested the utility of the resulting groupings for over-representation analysis. RESULTS: We successfully mapped 85% of RCV terms to logical OWL definitions and showed that these could be used to recapitulate and extend manual mappings between RCV terms and the sets of GO terms subsumed by them. We also show that gene sets derived from the resulting GO terms sets can be used to detect the signatures of cell and tissue types in whole genome expression data. CONCLUSIONS: The rich formal structure of the GO makes it possible to use reasoning to dynamically generate novel, biologically relevant groupings of GO terms. GO term groupings generated with this approach can be used in. over-representation analysis to detect cell and tissue type signatures in whole genome expression data.

Gradients and thresholds: BMP response gradients unveiled in Drosophila embryos.

Bone morphogenetic proteins (BMP) direct dorsal-ventral patterning in both invertebrate and vertebrate embryos, with strong evolutionary conservation of molecular components of the pathway. Dorsal-ventral patterning of the early Drosophila embryo is a powerful experimental system to probe mechanisms of BMP gradient formation and interpretation. Recent studies have found that spatial patterns of activated BMP signal transducers in Drosophila go through an unexpected transition: a shallow gradient of weak responses at mid-cellularization changes to a step gradient of stronger responses in cellularized embryos. The transition between two gradients of different shape yields new insights into the progression of Drosophila dorsal-ventral patterning and raises new issues about the mechanisms of gradient formation.

Stepwise formation of a SMAD activity gradient during dorsal-ventral patterning of the Drosophila embryo.

Genetic evidence suggests that the Drosophila ectoderm is patterned by a spatial gradient of bone morphogenetic protein (BMP). Here we compare patterns of two related cellular responses, both signal-dependent phosphorylation of the BMP-regulated R-SMAD, MAD, and signal-dependent changes in levels and sub-cellular distribution of the co-SMAD Medea. Our data demonstrate that nuclear accumulation of the co-SMAD Medea requires a BMP signal during blastoderm and gastrula stages. During this period, nuclear co-SMAD responses occur in three distinct patterns. At the end of blastoderm, a broad dorsal domain of weak SMAD response is detected. During early gastrulation, this domain narrows to a thin stripe of strong SMAD response at the dorsal midline. SMAD response levels continue to rise in the dorsal midline region during gastrulation, and flanking plateaus of weak responses are detected in dorsolateral cells. Thus, the thresholds for gene expression responses are implicit in the levels of SMAD responses during gastrulation. Both BMP ligands, DPP and Screw, are required for nuclear co-SMAD responses during these stages. The BMP antagonist Short gastrulation (SOG) is required to elevate peak responses at the dorsal midline as well as to depress responses in dorsolateral cells. The midline SMAD response gradient can form in embryos with reduced dpp gene dosage, but the peak level is reduced. These data support a model in which weak BMP activity during blastoderm defines the boundary between ventral neurogenic ectoderm and dorsal ectoderm. Subsequently, BMP activity creates a step gradient of SMAD responses that patterns the amnioserosa and dorsomedial ectoderm.