Manabu Ishii

PosMed (Positional Medline): prioritizing genes with an artificial neural network comprising medical documents to accelerate positional cloning

PosMed (http://omicspace.riken.jp/) prioritizes candidate genes for positional cloning by employing our original database search engine GRASE, which uses an inferential process similar to an artificial neural network comprising documental neurons (or ‘documentrons’) that represent each document contained in databases such as MEDLINE and OMIM. Given a user-specified query, PosMed initially performs a full-text search of each documentron in the first-layer artificial neurons and then calculates the statistical significance of the connections between the hit documentrons and the second-layer artificial neurons representing each gene. When a chromosomal interval(s) is specified, PosMed explores the second-layer and third-layer artificial neurons representing genes within the chromosomal interval by evaluating the combined significance of the connections from the hit documentrons to the genes. PosMed is, therefore, a powerful tool that immediately ranks the candidate genes by connecting phenotypic keywords to the genes through connections representing not only gene–gene interactions but also other biological interactions (e.g. metabolite–gene, mutant mouse–gene, drug–gene, disease–gene and protein–protein interactions) and ortholog data. By utilizing orthologous connections, PosMed facilitates the ranking of human genes based on evidence found in other model species such as mouse. Currently, PosMed, an artificial superbrain that has learned a vast amount of biological knowledge ranging from genomes to phenomes (or ‘omic space’), supports the prioritization of positional candidate genes in humans, mouse, rat and Arabidopsis thaliana.

The RIKEN integrated database of mammals

The RIKEN integrated database of mammals (http://scinets.org/db/mammal) is the official undertaking to integrate its mammalian databases produced from multiple large-scale programs that have been promoted by the institute. The database integrates not only RIKEN’s original databases, such as FANTOM, the ENU mutagenesis program, the RIKEN Cerebellar Development Transcriptome Database and the Bioresource Database, but also imported data from public databases, such as Ensembl, MGI and biomedical ontologies. Our integrated database has been implemented on the infrastructure of publication medium for databases, termed SciNetS/SciNeS, or the Scientists’ Networking System, where the data and metadata are structured as a semantic web and are downloadable in various standardized formats. The top-level ontology-based implementation of mammal-related data directly integrates the representative knowledge and individual data records in existing databases to ensure advanced cross-database searches and reduced unevenness of the data management operations. Through the development of this database, we propose a novel methodology for the development of standardized comprehensive management of heterogeneous data sets in multiple databases to improve the sustainability, accessibility, utility and publicity of the data of biomedical information.

PosMed-plus: an intelligent search engine that inferentially integrates cross-species information resources for molecular breeding of plants.

Molecular breeding of crops is an efficient way to upgrade plant functions useful to mankind. A key step is forward genetics or positional cloning to identify the genes that confer useful functions. In order to accelerate the whole research process, we have developed an integrated database system powered by an intelligent data-retrieval engine termed PosMed-plus (Positional Medline for plant upgrading science), allowing us to prioritize highly promising candidate genes in a given chromosomal interval(s) of Arabidopsis thaliana and rice, Oryza sativa. By inferentially integrating cross-species information resources including genomes, transcriptomes, proteomes, localizomes, phenomes and literature, the system compares a users query, such as phenotypic or functional keywords, with the literature associated with the relevant genes located within the interval. By utilizing orthologous and paralogous correspondences, PosMed-plus efficiently integrates cross-species information to facilitate the ranking of rice candidate genes based on evidence from other model species such as Arabidopsis. PosMed-plus is a plant science version of the PosMed system widely used by mammalian researchers, and provides both a powerful integrative search function and a rich integrative display of the integrated databases. PosMed-plus is the first cross-species integrated database that inferentially prioritizes candidate genes for forward genetics approaches in plant science, and will be expanded for wider use in plant upgrading in many species.

ARTADE2DB: Improved Statistical Inferences for Arabidopsis Gene Functions and Structure Predictions by Dynamic Structure-Based Dynamic Expression (DSDE) Analyses

Recent advances in technologies for observing high-resolution genomic activities, such as whole-genome tiling arrays and high-throughput sequencers, provide detailed information for understanding genome functions. However, the functions of 50% of known Arabidopsis thaliana genes remain unknown or are annotated only on the basis of static analyses such as protein motifs or similarities. In this paper, we describe dynamic structure-based dynamic expression (DSDE) analysis, which sequentially predicts both structural and functional features of transcripts. We show that DSDE analysis inferred gene functions 12% more precisely than static structure-based dynamic expression (SSDE) analysis or conventional co-expression analysis based on previously determined gene structures of A. thaliana. This result suggests that more precise structural information than the fixed conventional annotated structures is crucial for co-expression analysis in systems biology of transcriptional regulation and dynamics. Our DSDE method, ARabidopsis Tiling-Array-based Detection of Exons version 2 and over-representation analysis (ARTADE2-ORA), precisely predicts each gene structure by combining two statistical analyses: a probe-wise co-expression analysis of multiple transcriptome measurements and a Markov model analysis of genome sequences. ARTADE2-ORA successfully identified the true functions of about 90% of functionally annotated genes, inferred the functions of 98% of functionally unknown genes and predicted 1,489 new gene structures and functions. We developed a database ARTADE2DB that integrates not only the information predicted by ARTADE2-ORA but also annotations and other functional information, such as phenotypes and literature citations, and is expected to contribute to the study of the functional genomics of A. thaliana. URL: http://artade.org.

PosMed: ranking genes and bioresources based on Semantic Web Association Study

Positional MEDLINE (PosMed; http://biolod.org/PosMed) is a powerful Semantic Web Association Study engine that ranks biomedical resources such as genes, metabolites, diseases and drugs, based on the statistical significance of associations between user-specified phenotypic keywords and resources connected directly or inferentially through a Semantic Web of biological databases such as MEDLINE, OMIM, pathways, co-expressions, molecular interactions and ontology terms. Since 2005, PosMed has long been used for in silico positional cloning studies to infer candidate disease-responsible genes existing within chromosomal intervals. PosMed is redesigned as a workbench to discover possible functional interpretations for numerous genetic variants found from exome sequencing of human disease samples. We also show that the association search engine enhances the value of mouse bioresources because most knockout mouse resources have no phenotypic annotation, but can be associated inferentially to phenotypes via genes and biomedical documents. For this purpose, we established text-mining rules to the biomedical documents by careful human curation work, and created a huge amount of correct linking between genes and documents. PosMed associates any phenotypic keyword to mouse resources with 20 public databases and four original data sets as of May 2013.

OmicBrowse: a Flash-based high-performance graphics interface for genomic resources

OmicBrowse is a genome browser designed as a scalable system for maintaining numerous genome annotation datasets. It is an open source tool capable of regulating multiple user data access to each dataset to allow multiple users to have their own integrative view of both their unpublished and published datasets, so that the maintenance costs related to supplying each collaborator exclusively with their own private data are significantly reduced. OmicBrowse supports DAS1 imports and exports of annotations to Internet site servers worldwide. We also provide a data-download named OmicDownload server that interactively selects datasets and filters the data on the selected datasets. Our OmicBrowse server has been freely available at http://omicspace.riken.jp/ since its launch in 2003. The OmicBrowse source code is downloadable from http://sourceforge.net/projects/omicbrowse/.

Logical Operation Based Literature Association with Genes and its application, PosMed.

Makita Y, Kobayashi N, Mochizuki Y, Yoshida Y, Asano S, Heida N, Deshpande M, Bhatia R, Matsushima A, Ishii M, Kawaguchi S, Iida K, Hanada K, Kuromori T, Seki M, Shinozaki K, Toyoda T. PosMed-plus: an intelligent search engine that inferentially integrates cross-species information resources for molecular breeding of plants. Plant Cell Physiol. 2009 50(7):1249-59. PMID: 19528193 Yoshida Y, Makita Y, Heida N, Asano S, Matsushima A, Ishii M, Mochizuki Y, Masuya H, Wakana S, Kobayashi N, Toyoda T. PosMed (Positional Medline): prioritizing genes with an artificial neural network comprising medical documents to accelerate positional cloning. Nucleic Acids Res. 2009 37(Web Server issue):W147-52. PMID: 19468046 PosMed prioritizes candidate genes for positional cloning by employing our original database search engine GRASE, which uses an inferential process similar to an artificial neural network comprising documental neurons (or ‘documentrons’) that represent each document contained in databases such as MEDLINE and OMIM (Yoshida, et al. 2009). PosMed immediately ranks the candidate genes by connecting phenotypic keywords to the genes through connections representing gene–gene interactions other biological relationships, such as metabolite–gene, mutant mouse– gene, drug–gene, disease–gene, and protein–protein interactions, ortholog data, and gene–literature connections. To make proper relationships between genes and literature, we manually curate queries, which are defined by logical operation rules, against MEDLINE. For example, to detect a set of MEDLINE documents for the AT1G03880 gene in A. thaliana, we applied the following logical query: (‘AT1G03880’ OR ‘CRU2’ OR ‘CRB’ OR ‘CRUCIFERIN 2’ OR ‘CRUCIFERIN B’) AND (‘Arabidopsis’) NOT (‘chloroplast RNA binding’). Curators refined these queries in mouse, rice and A. thaliana. For human and rat genes, we use mouse curation results via ortholog genes in PosMed.