• See-Kiong Ng (NTU & I2R, Singapore)
  •    
    Dissecting protein interaction networks with meso-scale network motifs
    Recent works in network analysis have revealed the existence of network motifs in biological networks such as the protein-protein interaction (PPI) networks. However, existing motif mining algorithms are not sufficiently scalable to find mesoscale network motifs. Also, there has been little or no work to systematically exploit the extracted network motifs for dissecting the vast interactomes. In this talk, we describe our efficient network motif discovery algorithm, NeMoFinder, that can mine meso-scale network motifs that are repeated and unique in large PPI networks. Using NeMoFinder, we successfully discovered, for the first time, up to size-12 network motifs in a large whole-genome S. cerevisiae (Yeast) PPI network. We also show that such network motifs can be systematically exploited for indexing the reliability of PPI data that were generated via highly erroneous high-throughput experimental methods.

  • Hwee Tou Ng (NUS, SoC, Singapore)
  •    
    Who Did What to Whom: Broad-Coverage Semantic Role Labeling of Free Text
    There is much recent research on broad-coverage semantic role labeling of free text, fueled by the availability of large annotated corpora such as PropBank, FrameNet, and NomBank. In this growing body of work, the task is to identify the semantic role of constituents in sentences from any unrestricted free text. The semantic representation produced can be useful in applications such as information extraction and question answering. In this talk, I will present our recent research on improving the accuracy of semantic role labeling of verbs, by exploiting the surrounding semantic arguments in a sentence. I will also present our initial work on semantic role labeling of nouns based on the NomBank corpus.

  • Jinah Park (ICU, Korea)
  •    
    Force-directed relation network layout on a layered graph for visualizing unknown relations
    As most relations in protein-protein interaction networks have been discovered one by one empirically, discovering new relations is a challenging and usually time consuming job. If we know that some objects have similar functions, however, we can make inferences on the relationship among such objects more readily, and these inferences can avoid many false trials and errors in the discovery process. An ontology is a structured representation of conceptual knowledge. This hierarchical knowledge can be applied for the inference of relations among objects since objects with similar functions share similar ontology terms. Therefore, combining the relation network with an ontology makes it possible to reflect this kind of knowledge and we can infer some unknown relations as well. We propose a visualization method in 3D space, to examine a specific relation network based on a proper ontology structure. In our visualization system, we add a degree of freedom to the conventional layered drawing algorithm so that the position of the term in an ontology tree can move like a mobile. And we combine it with a modified spring embedder model to map the relation network onto the ontology tree. We have used protein-protein interaction data from Ubiquitination Information System for the relation network, and Gene Ontology for the ontology structure. The proposed method lays out the protein relation data in 3D space with a meaningful distance measure.

  • Gwan-Su Yi (ICU, Korea; with Choong-Hyun Sun)
  •    
    Ontology Driven Database Model for Mitochondrial SNP Study
    Simple database structure found in the most SNP analysis systems is hard to integrate complex biological information. It is necessary to employ ontology-driven information system to integrate various level of biological information. Currently, some ontology-driven databases have been introduced in this filed, but the benefit of ontology system is not fully implemented. The well-known biological ontology system like Gene Ontology (GO) includes only limited number of properties of current ontology technology. Semantic information across complex relation chains among data can be retrieved efficiently by the description of full properties of each term and by the analysis of relationship among data with appropriate reasoning. In this way, the domain-specific knowledge development from given information can be maximized. In this research, we propose a model of ontology-based information system equipped with Web Ontology Language and a reasoning engine for specifically mitochondrial SNP analysis. The proposed system supports the finding of SNP-gene-disease relations, automatic data validity and knowledge conflict checking. This system will be able to provide more comprehensive information to the users by increasing the semantic relationships and their reasoning outputs.

  • Choe, Jae-Woong (Korea University, Korea)
  •    
    Treating Classifiers in Database Semantics
    Text information on the internet would include a lot of quantity related expressions that require proper handling in any NLP search system. Especially, languages that make use of classifiers, for example, Korean, Japanese, or many others over the world, need some special attention in the sense that they show remarkable diversity in the construction of the noun phrases. In this presentation, we provide an analysis of various constructions of noun phrases involving classifiers in Korean, using the time-linear derivations of Database Semantics (Hausser 1999, 2006). We first show that there are at least 28 possible noun phrase constructions or templates in Korean that can be created out of head nouns and optional numerals, classifiers/quantifiers or case markers. We then present how those constructions can be derived step by step on the basis of three left-associative rules. The derivations are given in their implemented form in JSL! IM (Hausser 2006).