Biological knowledge is widely represented in the form of ontologies and ontology-based annotations. Biomedical ontologies describe known phenomena in biology using formal axioms, and the annotations associate an entity (e.g. genes, diseases, chemicals, etc.) with a set of biological concepts. In addition to formally structured axioms, ontologies contain meta-data in the form of annotation properties expressed mostly in natural language which provide valuable pieces of information that characterize ontology concepts. The structure and information contained in ontologies and their annotations make them valuable for use in machine learning, data analysis and knowledge extraction tasks. I develop the first approaches that can exploit all of the information encoded in ontologies, both formal and informal, to learn feature embeddings of biological concepts and biological entities based on their annotations to ontologies. Notably, I develop the first approach to use all the formal content of ontologies in the form of logical axioms and entity annotations to generate feature vectors of biological entities using neural language models. I extend the proposed algorithm by enriching the obtained feature vectors through representing the natural language annotation properties within the ontology meta-data as axioms. Transfer learning is then applied to learn from the biomedical literature and apply on the formal knowledge of ontologies. To optimize learning that combines the formal content of biomedical ontologies and natural language data such as the literature, I also propose a new approach that uses self-normalization with a deep Siamese neural network that improves learning from both the formal knowledge within ontologies and textual data. I validate the proposed algorithms by applying them to the Gene Ontology to generate feature vectors of proteins based on their functions, and to the PhenomeNet ontology to generate features of genes and diseases based on the phenotypes they are associated with. The generated features are then used to train a variety of machinelearning based classifiers to perform different prediction tasks including the prediction of protein interactions, gene–disease associations and the toxicological effects of chemicals. I also use the proposed methods to conduct the first quantitative evaluation of the quality of the axioms and meta-data included in ontologies to prove that including axioms as background improves ontology-based prediction. The proposed approaches can be applied to a wide range of other bioinformatics research problems including similarity-based prediction and classification of interaction types using supervised learning, or clustering.

Biological knowledge is widely represented in the form of ontologies and ontologybased annotations. Biomedical ontologies describe known phenomena in biology using formal axioms, and the annotations associate an entity (e.g. genes, diseases, chemicals, etc.) with a set of biological concepts. In addition to formally structured axioms, ontologies contain meta-data in the form of annotation properties expressed mostly in natural language which provide valuable pieces of information that characterize ontology concepts. The structure and information contained in ontologies and their annotations make them valuable for use in machine learning, data analysis and knowledge extraction tasks. I develop the rst approaches that can exploit all of the information encoded in ontologies, both formal and informal, to learn feature embeddings of biological concepts and biological entities based on their annotations to ontologies. Notably, I develop the rst approach to use all the formal content of ontologies in the form of logical axioms and entity annotations to generate feature vectors of biological entities using neural language models. I extend the proposed algorithm by enriching the obtained feature vectors through representing the natural language annotation properties within the ontology meta-data as axioms. Transfer learning is then applied to learn from the biomedical literature and apply on the formal knowledge of ontologies. To optimize learning that combines the formal content of biomedical ontologies and natural language data such as the literature, I also propose a new approach that uses self-normalization with a deep Siamese neural network that improves learning from both the formal knowledge within ontologies and textual data. I validate the proposed algorithms by applying them to the Gene Ontology to generate feature vectors of proteins based on their functions, and to the PhenomeNet ontology to generate features of genes and diseases based on the phenotypes they are associated with. The generated features are then used to train a variety of machinelearning based classi ers to perform di erent prediction tasks including the prediction of protein interactions, gene{disease associations and the toxicological e ects of chemicals. I also use the proposed methods to conduct the rst quantitative evaluation of the quality of the axioms and meta-data included in ontologies to prove that including axioms as background improves ontology-based prediction. The proposed approaches can be applied to a wide range of other bioinformatics research problems including similarity-based prediction and classi cation of interaction types using supervised learning, or clustering.

Ontologies are a key technology for the Semantic Web. In different areas, a large number of ontologies have been developed so far by different people or organizations under the same domains and many of them contain overlapping information. In order to get more benefit from different ontologies having inter-related knowledge they have to be aligned or merged. A number of systems have been developed for aligning and merging ontologies and various alignment strategies are used in these systems. However, there is no system available which supports multiple alignment sessions for aligning large ontologies adequately. In this thesis work we propose a session-based framework for aligning and merging large ontologies. We have implemented two types of sessions, computation sessions to generate suggestions and validation sessions to validate these generated suggestions. Furthermore after categorizing suggestions into accepted and rejected ones, we generated partial reference alignment (PRA) that can be used to compute similarities between terms and to filter mapping suggestions. We have also proposed recommendation process integrated with computation and validation sessions in order to find out which matchers, and combinations are better to use for alignment process. Either computation and validation sessions may use the recommended settings or the user can select other matchers and combinations.

Animal classification needs vary by use and application. The Linnaean taxonomy is an important animal classification scheme but does not portray key animal identifying information like sex, age group, physiologic stage, living environment and role in production systems such as farms. Ontologies are created and used for defining, organizing and classifying information in a domain to enable learning and sharing of information. This work develops an ontology of animal classes that form the basis for communication of animal identifying information among animal managers, medical professionals caring for animals and biomedical researchers involved in disciplines as diverse as wildlife ecology and dairy science. The Animals in Context Ontology (ACO) was created from an extension and subset of the Systematized Nomenclature of Medicine — Clinical Terms (SNOMED-CT). The principles of the Open Biological and Biomedical Ontologies (OBO) Foundry were followed and freely available tools were used. ACO includes normal development and physiologic animal classes as well animal classes where humans have assigned the animal's role. ACO is interoperable with and includes classes from other OBO Foundry ontologies such as the Gene Ontology (GO). Meeting many of the OBO Foundry principles was straightforward but difficulties were encountered with missing and problematic content in some of the OBO ontologies. Additions and corrections were submitted to four ontologies. Some information in ACO could not be represented formally because of inconsistency in husbandry practices. ACO classes are of interest to science, medicine and agriculture, and can connect information between animal and human systems to enable knowledge discovery. / Master of Science

En médecine personnalisée, les mesures et les descriptions radiologiques jouent un rôle important. En particulier, elles facilitent aux cliniciens l’établissement du diagnostic, la prise de décision thérapeutique ainsi que le suivi de la réponse au traitement. On peut citer à titre d’exemple, les critères d’évaluation RECIST (en anglais Response Evaluation Criteria in Solid Tumors). De nombreuses études de corrélation en radiologie-pathologie montrent que les caractéristiques d'imagerie quantitative et qualitative sont associées aux altérations génétiques et à l'expression des gènes. Par conséquent, une gestion appropriée des phénotypes d'imagerie est nécessaire pour faciliter leur utilisation et leur réutilisation dans de multiples études concernant les mesures radiologiques. En littérature, les mesures radiologiques qui caractérisent les processus biologiques des sujets imagés sont appelées biomarqueurs d'imagerie. L'objectif principal de cette thèse est de proposer une conceptualisation ontologique des biomarqueurs d'imagerie pour rendre leur sens explicite et formel, améliorer le reporting structuré des images. La première partie de la thèse présente une ontologie générique qui définit les aspects fondamentaux du concept de biomarqueur d'imagerie, à savoir : les caractéristiques biologiques mesurées, les protocoles de mesure et les rôles des biomarqueurs imagerie dans la prise de décision. La deuxième partie de la thèse traite des problèmes de modélisation sémantique liés à la description des données d’observation en neuro-imagerie en utilisant les connaissances biomédicales existantes. Ainsi, elle propose des solutions ''pertinentes'' aux situations les plus typiques qui doivent être modélisées dans le glioblastome. / In personalized medicine, radiological measurements and observations play an important role; in particular they help clinicians in making their diagnosis, selecting the appropriate treatment and monitoring the therapeutic response to an intervention as for example the Response Evaluation Criteria in Solid Tumors (RECIST). Many radiology-pathology correlation studies show that quantitative and qualitative imaging features are associated to genetic alterations and gene expression. Therefore, suitable management of imaging phenotypes is needed to facilitate their use and reuse in multiple studies regarding radiological measurements. In litterature, radiological measurements that characterize biological processes of imaged subjects are called imaging biomarkers. The main objective of this thesis is to propose an ontological conceptualisation of imaging biomarkers to make their meaning explicit and formal, improve structured reporting of images. The first part of the thesis presents a generic ontology that defines basic aspects of the imaging biomarker concept, namely; measured biological characteristic, measurement protocols and role in decision making application. The second part of the thesis adresses important semantic modeling challenges related to the description of neuro-imaging data using existing biomedical knowledge, as well as it proposes some “relevant” solutions to the most typical situations that need to be modeled in glioblastoma.

Ontologies provide an organization of language, in the form of a network or graph, which is amenable to computational analysis while remaining human-readable. Although they are used in a variety of disciplines, ontologies in the biomedical field, such as Gene Ontology, are of interest for their role in organizing terminology used to describe—among other concepts—the functions, locations, and processes of genes and gene-products. Due to the consistency and level of automation that ontologies provide for such annotations, methods for finding enriched biological terminology from a set of differentially identified genes in a tissue or cell sample have been developed to aid in the elucidation of disease pathology and unknown biochemical pathways. However, despite their immense utility, biomedical ontologies have significant limitations and caveats. One major issue is that gene annotation enrichment analyses often result in many redundant, individually enriched ontological terms that are highly specific and weakly justified by statistical significance. These large sets of weakly enriched terms are difficult to interpret without manually sorting into appropriate functional or descriptive categories. Also, relationships that organize the terminology within these ontologies do not contain descriptions of semantic scoping or scaling among terms. Therefore, there exists some ambiguity, which complicates the automation of categorizing terms to improve interpretability. We emphasize that existing methods enable the danger of producing incorrect mappings to categories as a result of these ambiguities, unless simplified and incomplete versions of these ontologies are used which omit problematic relations. Such ambiguities could have a significant impact on term categorization, as we have calculated upper boundary estimates of potential false categorizations as high as 121,579 for the misinterpretation of a single scoping relation, has_part, which accounts for approximately 18% of the total possible mappings between terms in the Gene Ontology. However, the omission of problematic relationships results in a significant loss of retrievable information. In the Gene Ontology, this accounts for a 6% reduction for the omission of a single relation. However, this percentage should increase drastically when considering all relations in an ontology. To address these issues, we have developed methods which categorize individual ontology terms into broad, biologically-related concepts to improve the interpretability and statistical significance of gene-annotation enrichment studies, meanwhile addressing the lack of semantic scoping and scaling descriptions among ontological relationships so that annotation enrichment analyses can be performed across a more complete representation of the ontological graph. We show that, when compared to similar term categorization methods, our method produces categorizations that match hand-curated ones with similar or better accuracy, while not requiring the user to compile lists of individual ontology term IDs. Furthermore, our handling of problematic relations produces a more complete representation of ontological information from a scoping perspective, and we demonstrate instances where medically-relevant terms--and by extension putative gene targets--are identified in our annotation enrichment results that would be otherwise missed when using traditional methods. Additionally, we observed a marginal, yet consistent improvement of statistical power in enrichment results when our methods were used, compared to traditional enrichment analyses that utilize ontological ancestors. Finally, using scalable and reproducible data workflow pipelines, we have applied our methods to several genomic, transcriptomic, and proteomic collaborative projects.

<p>Due to the explosion of the amount of biomedical data, knowledge and tools that are often publicly available over the Web, a number of difficulties are experienced by biomedical researchers. For instance, it is difficult to find, retrieve and integrate information that is relevant to their research tasks. Ontologies and the vision of a Semantic Web for life sciences alleviate these difficulties. In recent years many biomedical ontologies have been developed and many of these ontologies contain overlapping information. To be able to use multiple ontologies they have to be aligned or merged. A number of systems have been developed for aligning and merging ontologies and various alignment strategies are used in these systems. However, there are no general methods to support building such tools, and there exist very few evaluations of these strategies. In this thesis we give an overview of the existing systems. We propose a general framework for aligning and merging ontologies. Most existing systems can be seen as instantiations of this framework. Further, we develop SAMBO (System for Aligning and Merging Biomedical Ontologies) according to this framework. We implement different alignment strategies and their combinations, and evaluate them in terms of quality and processing time within SAMBO. We also compare SAMBO with two other systems. The work in this thesis is a first step towards a general framework that can be used for comparative evaluations of alignment strategies and their combinations.</p> / Report code: LiU-Tek-Lic-2006:6.

Using ontologies is a key technology in the semantic web. The semantic web helps people to store their data on the web, build vocabularies, and has written rules for handling these data and also helps the search engines to distinguish between the information they want to access in web easier. In order to use multiple ontologies created by different experts we need matchers to find the similar concepts in them to use it to merge these ontologies. Text based searches use the string similarity functions to find the equivalent concepts inside ontologies using their names.This is the method that is used in lexical matchers. But a global standard for naming the concepts in different research area does not exist or has not been used. The same name may refer to different concepts while different names may describe the same concept. To solve this problem we can use another approach for calculating the similarity value between concepts which is used in structural and constraint-based matchers. It uses relations between concepts, synonyms and other information that are stored in the ontologies. Another category for matchers is instance-based that uses additional information like documents related to the concepts of ontologies, the corpus, to calculate the similarity value for the concepts. Decision trees in the area of data mining are used for different kind of classification for different purposes. Using decision trees in an instance-based matcher is the main concept of this thesis. The results of this implemented matcher using the C4.5 algorithm are discussed. The matcher is also compared to other matchers. It also is used for combination with other matchers to get a better result.

The amount of biomedical information that is disseminated over the Web increases every day. This rich resource is used to find solutions to challenges across the life sciences. The Semantic Web for life sciences shows promise for effectively and efficiently locating, integrating, querying and inferring related information that is needed in daily biomedical research. One of the key technologies in the Semantic Web is ontologies, which furnish the semantics of the Semantic Web. A large number of biomedical ontologies have been developed. Many of these ontologies contain overlapping information, but it is unlikely that eventually there will be one single set of standard ontologies to which everyone will conform. Therefore, applications often need to deal with multiple overlapping ontologies, but the heterogeneity of ontologies hampers interoperability between different ontologies. Aligning ontologies, i.e. identifying relationships between different ontologies, aims to overcome this problem. A number of ontology alignment systems have been developed. In these systems various techniques and ideas have been proposed to facilitate identification of alignments between ontologies. However, there still is a range of issues to be addressed when we have alignment problems at hand. The work in this thesis contributes to three different aspects of identification of high quality alignments: 1) Ontology alignment strategies and systems. We surveyed the existing ontology alignment systems, and proposed a general ontology alignment framework. Most existing systems can be seen as instantiations of the framework. Also, we developed a system for aligning biomedical ontologies (SAMBO) according to this framework. We implemented various alignment strategies in the system. 2) Evaluation of ontology alignment strategies. We developed and implemented the KitAMO framework for comparative evaluation of different alignment strategies, and we evaluated different alignment strategies using the implementation. 3) Recommending optimal alignment strategies for different applications. We proposed a method for making recommendations.

Due to the explosion of the amount of biomedical data, knowledge and tools that are often publicly available over the Web, a number of difficulties are experienced by biomedical researchers. For instance, it is difficult to find, retrieve and integrate information that is relevant to their research tasks. Ontologies and the vision of a Semantic Web for life sciences alleviate these difficulties. In recent years many biomedical ontologies have been developed and many of these ontologies contain overlapping information. To be able to use multiple ontologies they have to be aligned or merged. A number of systems have been developed for aligning and merging ontologies and various alignment strategies are used in these systems. However, there are no general methods to support building such tools, and there exist very few evaluations of these strategies. In this thesis we give an overview of the existing systems. We propose a general framework for aligning and merging ontologies. Most existing systems can be seen as instantiations of this framework. Further, we develop SAMBO (System for Aligning and Merging Biomedical Ontologies) according to this framework. We implement different alignment strategies and their combinations, and evaluate them in terms of quality and processing time within SAMBO. We also compare SAMBO with two other systems. The work in this thesis is a first step towards a general framework that can be used for comparative evaluations of alignment strategies and their combinations. / <p>Report code: LiU-Tek-Lic-2006:6.</p>