A Nanopublication Framework for Biological Networks using Cytoscape.js

Abstract

-We leverage semantic technologies and Cytoscape.js to create a provenance-aware, probabilistic analysis platform for systems biology and evaluate its usefulness in discovering links between drugs and diseases. In our efforts to create a systematic approach to discovering new uses for existing drugs, we have developed Repurposing Drugs with Semantics (ReDrugS). ReDrugS is a data curation and publication framework that accepts data from nearly any database containing biological or chemical entity interactions and produces visualizations using Cytoscape.js. A semantic web service API is provided that enables search, traversal, and provides composite probabilities for the resulting graph of biological entities using the SADI web service framework and Nanopublications. We show how associations between a postive control, topiramate, allows us to independently reconstruct a positive control of epilepsy and migraine, and potential consequences on bone health.

I. INTRODUCTION

Drug repurposing can often lead to effective new treatments for diseases. The ReDrugs system we are developing can assist in this procedure through the integration of multiple systems biology, pharmacology, disease association, and gene expression databases into a coherent repository of individuallysupported assertions that can each be assigned their own probabilistic value. We have developed an initial database that includes drug/protein, protein/protein, and protein/biological process associations that is providing us a view into how drugs have the effects that they do.

II. METHODS

We deployed an instance of the RPI semantic web toolsuite, Prizms, at http://redrugs.tw.rpi.edu and the Comprehensive Knowledge Archive Network (CKAN) to http://data.melagrid.org to catalog the available datasets [ 1 ]. Cataloging is an ongoing process, but initial datasets were added to the catalog, initializing the Prizms conversion process. We were then able to use the Prizms infrastructure to generate RDF for publication to our SPARQL endpoint. We used the BigData RDF store with named graph and text indexing support enabled.

A. Inferring Probabilities

Two molecular biology and biochemistry experts, Michel Dumontier and Pascale Gaudet, assigned a score from low to high confidence of 1-3, evidence and/or technique associated with the interaction. The confidence measure was based on the comparative analyses of techniques [ 2 ], [ 3 ], and experience of the experts in reviewing data of this kind. The confidence assignment is based on a number of factors including degree of indirection in the assay, sensitivity and specificity of the approach, and reproducibility of results under different conditions. The confidence scores for both experts were encoded as classes of evidence, where each experimental method class was assigned two superclasses, one for each expert. This ontology was created from a spreadsheet and expanded to full inferences using Pellet [ 4 ]. At the same time, SPARQL-based reasoning is used to classify nanopublication assertions by their available evidence, and thereby assign a class of confidence codes to it.

B. SADI Web Service Interface

We developed four Semantic Automated Discovery and Integration (SADI) web services in Python1 to support easy access to the nanopublications. We use SADI to provide a discoverable, consistent API that can be re-used in other applications or directly consumed by analytical tools.

The services perform these computational tasks that would otherwise be difficult to perform with SPARQL queries. The services return only one interaction for each triple (source, interaction type, target) but multiple, probabilities per interaction, and more than one interaction per interaction type. This is because the interaction may have been recorded in multiple databases, based on different experimental methods. To provide a single probability score for each triple, the interactions are combined. This is done to indicate that multiple experiments that produce the same results reinforce each other, and should therefore give a higher overall probability than would be indicated by taking their mean.

P (x1...n) = CDF n X CDF 1 (P (xi))

! i=1 1For further information on developing web services in Python using SADI, see this tutorial: https://code.google.com/p/sadi/wiki/ BuildingServicesInPython

C. User Interface

Users can search for biological entities and processes, which can then be autocompleted to specific entities that are in the ReDrugS graph. Users can then add those entities and processes to the displayed graph and retrieve upstream and downstream connections and link out to more details for every entity. Cytoscape.js is used as the main rendering and network visualization tool, and provides node and edge rendering, layout, and network analysis capabilities.

III. EVALUATION

In order to evaluate this knowledge base, we developed a demonstration web interface2 based on the Cytoscape.js3. It lets users enter biological entity names, and as the user types, the text is resolved to a list of entities to be selected. After that, the entity is submitted to all three SADI services via a basic JavaScript SADI client.4 The resulting interactions and nodes are added to the Cytoscape.js graph, which can be laid out according to a number of algorithms. Users are also able to select nodes and populate upstream or downstream connections. An example of this is shown in Figure 1. This figure was obtained by putting “Topiramate” as a query in the search box, which returned all of the biological entities that topiramate is directly associated with. We then expanded the network downstream to see what biological entities are affected by topiramate’s targets.

We are able to successfully navigate a protein-drug-disease interaction graph that is a consensus of 16 diverse sources, to infer prior probabilities for more than three million individual assertions using their provenance and experts’ confidence in different experimental methods and to find drug/disease associations that are not directly expressed by any one database.

We plan to add further data sources, especially those that provide direct experimental results that predict protien (or gene)/disease associations like the Gene Expression Atlas (in progress) [ 5 ]. Further, we are very interested in integrating the newest version of the Connectivity Map dataset [ 6 ], as it provides gene expression signature similarities for a large number of chemical and genetic perturbations. Finally, as we develop new hypotheses about potential new drug effects, we plan to test them using a new three-dimensional cellular microarray to perform high throughput drug screening [ 7 ] with reference samples.

V. CONCLUSION

We have developed a framework for collecting, searching, analyzing, and visualizing important components of biological systems. We were able to build this by converting existing databases into a common nanopublication structure that uses the provenance of the database records to determine the quality of any given piece of information through the methods used to provide it. We use the Semantic Automated Discovery and Integration framework to provide simple access to data, and can visualize results using an existing interaction graph tool. The resulting application makes it easy to search for biological entities and see how they interact. We have already found some hypotheses of proteins through which drugs influence disease conditions. We plan to expand the loaded set of data with protein/disease associations as well as gene expression profiles, and will be using ReDrugS to produce prospective testable hypotheses.

ACKNOWLEDGMENTS

A special thanks to Pascale Gaudet, who, with Michel Dumontier, evaluated the experimental methods and evidence codes listed in the Protein/Protein Interaction Ontology and Gene Ontology.

Tall version (Preferred) Use this version in the majority of cases. 5AMwwSaoaslsuGtieoennnseeLrroaagtteeoddUBBsyayge Guidelines

G

XX aaMdI_ir0e4c0t7interaction hhasa-st-atragregte:t: SLSCL4CA48A8 hhasa-sp-apratrictiicpiapnatn:t:CAC2A2

References

[1] J. P. McCusker , T. Lebo , M. Krauthammer , and D. L. McGuinness , “ Next Generation Cancer Data Discovery, Access, and Integration Using Prizms and Nanopublications,” in Data Integration in the Life Sciences . Springer, 2013 , pp. 105 - 112 .

[2] J. C. Obenauer and M. B. Yaffe , “ Computational prediction of proteinprotein interactions,” in Protein-Protein Interactions . Springer, 2004 , pp. 445 - 467 .

[3] E. Sprinzak , S. Sattath , and H. Margalit , “ How reliable are experimental protein-protein interaction data?” Journal of molecular biology , vol. 327 , no. 5 , pp. 919 - 923 , 2003 .

[4] E. Sirin , B. Parsia , B. C. Grau , A. Kalyanpur , and Y. Katz , “ Pellet: A practical owl-dl reasoner,” Web Semantics: science , services and agents on the World Wide Web , vol. 5 , no. 2 , pp. 51 - 53 , 2007 .

[5] R. Petryszak , T. Burdett , B. Fiorelli , N. A. Fonseca , M. GonzalezPorta , E. Hastings, W. Huber , S. Jupp , M. Keays , N. Kryvych , and et al., “Expression Atlas update-a database of gene and transcript expression from microarray- and sequencing-based functional genomics experiments , ” Nucleic Acids Research , vol. 42 , no. D1 , p. D926 - D932 , Jan 2014 . [Online]. Available: http://dx.doi.org/10.1093/nar/gkt1270

[6] J. Lamb , E. D. Crawford , D. Peck , J. W. Modell , I. C. Blat , M. J. Wrobel , J. Lerner , J.-P. Brunet , A. Subramanian , K. N. Ross et al., “ The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease,” science , vol. 313 , no. 5795 , pp. 1929 - 1935 , 2006 .

[7] M.-Y. Lee , R. A. Kumar , S. M. Sukumaran , M. G. Hogg , D. S. Clark , and J. S. Dordick , “ Three-dimensional cellular microarray for high-throughput toxicology assays , ” Proceedings of the National Academy of Sciences , vol. 105 , no. 1 , pp. 59 - 63 , 2008 .