Immune Cell Ontology for Networks (ICON) Immunology Ontologies and Their Applications in Processing Clinical Data June 11-13, Buffalo, NY Confessions • • • • • • • I am an ontological newbie Idea for a new ontology of immune networks Immunologists I’ve talked to like the idea Biostatisticians I’ve talked to like the idea So, possibly not entirely stupid Looking for feedback and advice Looking for friendly collaborators Immunological case-control studies Normal Donor Lupus Patient Typical case-control study • Data collection – Hundreds of cell subsets from flow cytometry – Dozens of cytokines from Luminex – Other assays (IHC, single cell PCR etc) • Data analysis – Pairwise comparisons • Apply Bonferroni correction gives p >> 0.05 – Statistical aggregation e.g. PCA • Often difficult to give biological interpretation HOW DOES AN IMMUNE RESPONSE ACTUALLY WORK? Cancer microenvironment network Missing biological knowledge • Immune response does not consist of isolated cells and cytokines acting independently • Networks coordinated by cell-cell communication • Gap – immune network ontology • Applied ontology that draws strength from pre-existing ontologies Is a network ontology feasible? • Analysis of regulatory networks suggest that networks map to dynamical attractors • Typically surprisingly few attractors given potential combinatorial explosion • Examples – Boolean regulatory networks (e.g. Kauffman) – Recurring gene network motifs (e.g. Alon) What’s needed? • Networks consist of cells that communicate via contact- and cytokine-mediated signaling – Components • Cells, cell surface molecules, cytokines • Cell-cell interactions may be specific to particular species, local environments and disease states – Contexts • Species, tissue, disease Gene Ontology cell surface molecule Protein Ontology cytokine Cell Ontology cell Components network Contexts tissue FMA disease Disease Ontology species NCBI Taxonomy Tentative construction strategy • Iterate – Build cheap “weak links” graph database by text mining • Edges for cell:cell surface molecule, cell surface molecule:cell surface molecule, cell:cytokine, cytokine:cell surface molecule • Question: Does text mining work for anyone here? – Human review to identify spurious links and add species, disease and tissue contexts – Use “confirmed” and “spurious” links as training, validation and test data sets to improve text mining • Split into networks – Split into discrete subgraphs by cutting “weakest” links based on some method of assigning weights to edges Usage • Queries – Find networks associated with a disease – Find cell subsets, receptors and cytokines associated with a network – Find reagents associated with cell subsets, receptors and cytokines – Find networks most relevant for given cell subsets, receptors and cytokines • Applications – Reference, targeted assay development, better informed fishing expeditions – Basic science – validate novel links or networks App: Web accessible reference • No existing database • Literature review is laborious • Useful public resource App: Targeted assay development • What networks are potentially active in disease X? • Which are the most informative cell subsets and/or cytokines for these networks? • What reagents are available to identify the cell subsets and/or cytokines of interest? (Needs additional reagent database) App: Better fishing expeditions • Sets of cells +/- cytokines in networks • Test for enrichment of networks in treatment groups rather than pairwise-comparisons • Adapt statistical methods developed for enrichment analysis in expression array data (e.g. TANGO or GSEA) • Allows integration of immune biomarkers over multiple panels (e.g. T, B, innate flow panels, Luminex, immunohistochemistry) STARTING POINT T cell social network analysis Initial networks in Protégé (courtesy of Anna Maria) Acknowledgements • Duke Center for Computational Immunology – Tom Kepler – Lindsay Cowell – Anna Maria Masci • Duke Immune Profiling Cores – Kent Weinhold – David Murdoch – Janet Staats – Sarah Sparks