Malika Mahoui

Towards Reproducible eScience in the Cloud

Whether it be data from ubiquitous devices such as sensors or data generated from telescopes or other laboratory instruments, technology apparent in many scientific disciplines is generating data at rates never witnessed before. Computational scientists are among the many who perform inductive experiments and analyses on these data with the goal of answering scientific questions. These computationally demanding experiments and analyses have become a common occurrence, resulting in a shift in scientific discovery, and thus leading to the term eScience. To perform eScience experiments and analysis at scale, one must have an infrastructure with enough computing power and storage space. The advent of cloud computing has allowed infrastructures and platforms to be created with theoretical limitless bounds, thus providing an attractive solution to this need. In this work, we create a reproducible process for the construction of eScience computing environments on top of cloud computing infrastructures. Our solution separates the construction of these environments into two distinct layers: (1) the infrastructure layer and (2) the software layer. We provide results of running our framework on two different computational clusters within two separate cloud computing environments to demonstrate that our framework can facilitate the replication or extension of an eScience experiment.

A Dynamic Workflow Approach for the Integration of Bioinformatics Services

Modern biological and chemical studies rely on life science databases as well as sophisticated software tools (e.g., homology search tools, modeling and visualization tools). These tools often have to be combined and integrated in order to support a given study. SIBIOS (System for the Integration of Bioinformatics Services) serves this purpose. The services are both life science database search services and software tools. The task engine is the core component of SIBIOS. It supports the execution of dynamic workflows that incorporate multiple bioinformatics services. The architecture of SIBIOS, the approaches to addressing the heterogeneity as well as interoperability of bioinformatics services, including data integration are presented in this paper.

Knowledge Discovery and Data Mining of Free Text Radiology Reports

Medical Knowledge Discovery and Data Mining (KDD) over text is a promising yet difficult technology for unlocking meaning and uncovering associations in vast clinical text repositories. We report our experience in developing a new text analytic system called MEDAT or Medical Exploratory Data Analysis over Text, which overcomes several problems in text mining. The MEDAT system employs an annotated semantic index with a large number of assertions (propositions). The semantic index is able to capture complex assertions which encapsulate conceptual relationships including their modifiers at a granular level. The index represents semantically equivalent sentences with the same symbols, a necessary component for KDD semantic queries, including semantic Boolean and correlation queries. The graphical user interface enables users to perform complex semantic analysis of the Roentgen corpus, consisting of 594,000 de-identified radiology reports with 4.3 million sentences, without having to learn a programming language. The MEDAT architecture offers a novel framework for text mining in other medical domains.

SIBIOS: a system for the integration of bioinformatics services

Modern biological and chemical studies rely on life science databases as well as sophisticated software tools (e.g., homology search tools, modeling and visualization tools). These tools often have to be combined and integrated in order to support a given study. SIBIOS (system for the integration of bioinformatics services) serves this purpose. The services are both life science database search services and software tools. The task engine is the core component of SIBIOS. It supports the execution of dynamic workflows that incorporate multiple bioinformatics services. The architecture of SIBIOS, the approaches used to address the heterogeneity as well as interoperability of bioinformatics services, including data integration are presented in this paper.

A service discovery approach in support of Web service integration

Modern biological studies rely on life science Web databases as well as sophisticated Web-based software tools (e.g., homology search tools, modeling and visualization tools). These tools or Web services often have to be combined and integrated in order to support a given study (i.e. an in-silico experiment). The large number of the available Web services makes a service discovery process that can identify the set of services that satisfy a given number of constraints from the pool of all the available Web services crucial. Without such an automated service discovery process the user will have to have a thorough knowledge of all of the existing Web services, a task-which is impractical in the life science domain. In this paper, a scalable approach to service discovery in the biological domain is presented. The approach is based on and guided by a domain ontology.

SIBIOS ontology: a robust package for the integration and pipelining of bioinformatics services

The recent technological advancements in biological research have allowed researchers to advance their knowledge of the domain far beyond expectations. The advent of easily accessible biological web databases such as NCBI databases and associated tools such as BLAST are key components to this development. However, with the growing number of these web based biological research tools and data sources, the time necessary to invest in becoming a domain expert is immense. Therefore, it is important to allow for easy user deployment of the wealth of available data sources and tools necessary to conduct biological research. In this paper we discuss an approach to create and maintain a robust ontology knowledge base that serves as the core for SIBIOS, a workflow based integration system for bioinformatics tools and data sources. Further, deployment of the ontology in various components of SIBIOS is discussed.

An automated data utility clustering methodology using data constraint rules

Many data privacy models have been created in the last few years using the k-anonymization methodology including l-diversity, p-sensitive k-anonymity, and t-closeness. While these methods differ in their approaches and quality of the results, they all focus on ensuring the anonymization of the data while at the same time attempt to protect the quality of the data by minimizing the loss of the information contained in the original data set. In this paper, we propose an automated k-anonymity approach that uses clustering to maximize the utility of the data while ensuring that the data privacy is maintained. Our method employs data constraint rules, which are defined by the data research expert to represent especially informative distributions in categorical attributes or inflections points in a continuous attribute. The values of the data constraints are an integral component of our utility function, which is used to maximize the utility of the anonymized dataset. Finally, we present our experimental results that show that our approach meets or exceeds existing methods that do not incorporate data constraint rules.

Leveraging User Search Behavior to Design Personalized Browsing Interfaces for Healthcare Web Sites

Understanding and leveraging user search behavior is increasingly becoming a key component towards improving web sites functionality for the health care consumer and provider. In this study we propose to leverage user search behavior to design user-tailored browsing interfaces to better support locating information in healthcare websites at the point-of-need.

Semantic correspondence in federated life science data integration systems

For execution of complex biological queries, data integration systems often use several intermediate data sources because the domain coverage of individual sources is limited. Quality of intermediate sources differs greatly based on the method used for curation, frequency of updates and breadth of domain coverage, which affects the quality of the results. Therefore, integration systems should provide data provenance; i.e. information about the path used to obtain every record in the result. Furthermore, since query capabilities of web-accessible sources are limited, integration systems need to support refinement queries of finer granularity issued over the integrated data. However, unlike the individual sources, integration systems have to handle the absence of data and conflicts in the integrated data caused by inconsistencies among the sources. This paper describes the solution proposed by BACIIS, the Biological and Chemical Information Integration System, for providing data provenance and for supporting refinement queries over integrated data. Semantic correspondence between records from different sources is defined based on the links connecting these data sources including cross-references. Two characteristics of semantic correspondence, namely degree and cardinality, are identified based on the closeness of the links that exist between data records and based on the mappings between domains of data records respectively. An algorithm based on semantic correspondence is presented to handle absence of data and conflicts in the integrated data.

Exploring the usability of folksonomies in the online art museum community

This paper presents a usability evaluation of the Indianapolis Museum of Art website - as a typical art museum website supporting both tag-based search and user tagging of artwork - in an effort to explore how users access artwork while interacting with the museum online search and retrieval system. The usability study examined the extent of usage of Steve Tagger capabilities (annotation and use of tags in the process of searching/accessing artwork resources) deployed on the website. The usability test results showed that 55% of the users were able to successfully locate information on the website using both traditional searching techniques and folksonomies. However, only 34% of the users were able to successfully locate artwork using tags only. On the other hand, 95% of the participants were able to annotate an object by adding a term or tag to describe the artwork.