Publicação

Semantic perspectives for learning over biomedical knowledge graphs

Ver documento

Detalhes bibliográficos
Resumo:Knowledge graphs represent an unparalleled opportunity for machine learning in the biomedical domain, given their ability to enrich data with meaningful context through semantic representations, such as knowledge graph embeddings and semantic similarity. However, the specificity of many biomedical tasks contrasts with the broad domains covered by large and successful biomedical knowledge graphs that describe entities according to several perspectives — semantic aspects. This is particularly challenging for predicting specific relations between entities described in the knowledge graph when the graph itself does not encode these relations. Current semantic representation methods consider the knowledge graph as a whole, ignoring the different semantic aspects. This thesis hypothesizes that semantic representations that are able to distinguish semantic aspects can improve the performance and explainability of biomedical relation prediction tasks. This work investigated different paradigms for defining semantic aspects based on classes and properties and developed multiple semantic representation techniques for both individual entities and entity pairs, with a focus on their explainability. Extensive experiments in proteinprotein interaction and gene-disease association predictions supported the empirical evaluation of the proposed methods and demonstrated that semantic aspect-oriented representations improve both predictive performance and explainability, fostering biomedical research. This work further highlights that in complex and multi-disciplinary domains, where a single knowledge graph is used to support a wide variety of tasks, it is essential to shift from viewing knowledge graphs as a whole to focusing on specific semantic perspectives.
Autores principais:Sousa, Rita Isabel Torres de
Assunto:Knowledge graph Semantic similarity Knowledge graph embedding Machine learning Biomedical application Grafo de conhecimento Semelhança semântica Embedding de grafos de conhecimento Aprendizagem automática Aplicação biomédica
Ano:2024
País:Portugal
Tipo de documento:tese de doutoramento
Tipo de acesso:acesso aberto
Instituição associada:Universidade de Lisboa
Idioma:inglês
Origem:Repositório da Universidade de Lisboa

Registos relacionados

Registos relacionados