Formal axioms in biomedical ontologies improve analysis and interpretation of associated data

Abstract

Motivation:There are now over 500 ontologies in the life sciences. Over the past years, significant resources have been invested into formalizing these biomedical ontologies. Formal axioms in ontologies have been developed and used to detect and ensure ontology consistency, find unsatisfiable classes, improve interoperability, guide ontology extension through the application of axiom-based design patterns, and encode domain background knowledge. At the same time, ontologies have extended their amount of human-readable information such as labels and definitions as well as other meta-data. As a consequence, biomedical ontologies now form large formalized domain knowledge bases and have a potential to improve ontology-based data analysis by providing background knowledge and relations between biological entities that are not otherwise connected. Results: We evaluate the contribution of formal axioms and ontology meta-data to the ontology-based prediction of protein-protein interactions and gene-disease associations. We find that the formal axioms that have been created for the Gene Ontology and several other ontologies significantly improve ontologybased prediction models through provision of domain-specific background knowledge. Furthermore, we find that the labels, synonyms and definitions in ontologies can also provide background knowledge that may be exploited for prediction. The axioms and meta-data of different ontologies contribute in varying degrees to improving data analysis. Our results have major implications on the further development of formal knowledge bases and ontologies in the life sciences, in particular as machine learning methods are more frequently being applied. Our findings clearly motivate the need for further development, and the systematic, application-driven evaluation and improvement, of formal axioms in ontologies. Availability: https://

Self-normalizing learning on biomedical ontologies using a deep Siamese neural network

“…Embedding the classes, relations, and instances in ontologies can provide useful features for predictive models that rely on background knowledge, and these embeddings can incorporate ontology axioms as well as natural language annotations such as labels and definitions (Kulmanov et al, 2019;Liu-Wei et al, 2019;Althubaiti et al, 2019;Smaili et al, 2018a). However, using the natural language information in ontologies can also add noise, in particular when labels or descriptions use complex terms, such as chemical formulas, which are not easy to recognize in natural language text (Smaili et al, 2019). We propose a novel method that more closely integrates ontologies and natural language text, including both literature and the labels, definitions, or synonyms contained within ontologies themselves.…”

Predicting candidate genes from phenotypes, functions, and anatomical site of expression

“…Using ontologies and the background knowledge they contain in machine learning models can significantly improve their performance (Smaili et al, 2019a). Here, we developed an ontology-based machine learning method to prioritize candidate genes based on abnormal phenotypes observed in mouse models, the normal functions of gene products, and anatomical location of gene expression.…”

Ontology-based prediction of cancer driver genes

“…Determining 9 which mutations or genes act as a bottleneck in the generation of cancer is fraught with 10 problems, as cells carrying one or more driver mutations will also carry a large set of 11 co-selected, "passenger", mutations which constitute most of the normal somatic 12 mutation-load of the expanded cancer cell but which do not directly generate the 13 neoplastic phenotype [3]. 14 Much effort has gone into developing algorithms to identify driver genes and their 15 mutations, most of which are based on the frequency or pattern of mutations in 16 multiple tumors and their predicted pathogenicity. The goal of identifying cancer 17 drivers may be achieved at the level of gene, protein or pathways, and multiple 18 approaches have been attempted to date [4].…”

“…Over the past decades, a tightly integrated system of ontologies 52 has been developed that interlinks the knowledge about basic biological phenomena 53 through the use of logical axioms [13]. Exploring the information in this system of 54 ontologies can enable novel types of analysis [14] and the background knowledge in the 55 ontologies has the potential to significantly improve biomedical data analysis [15]. 56 We have developed a method that uses biological background knowledge about the 57 relation between genes or variants and their phenotypes, either on the cellular or whole 58 body organism level, as well as gene functions and cellular locations, to predict driver 59 genes and mutations.…”

“…193 Our method can predict cancer driver genes using only public background knowledge 194 about gene functions, cellular and organism phenotypes. The key novelty of our 195 algorithm is the ability to encode biological background knowledge in ontologies [15], 196 and therefore exploit inter-ontology links in the form of axioms; the predictive 197 performance of our method is best when combining different -yet related -ontologies. 198 While many of the axioms that relate classes in different ontologies have been created to 199 support ontology development and maintenance, we show that ontologies in the 200 biomedical domain now form a comprehensive web of biological background knowledge 201 that can -when exploited through appropriate learning algorithms -significantly connection between different ontologies is the Relation Ontology [25] which is central to 204 ontologies that make up the OBO Foundry [13].…”