A carregar...
Publicação
Boosting graph neural networks with knowledge for personalized medicine
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 565.38 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Cancer is one of the leading causes of death around the globe. Personalized medicine research
has been increasing in popularity in recent years due to technological advances such as Machine
Learning. However, clinical datasets often have a high number of features, requiring complex
processing for ML such as feature extraction, and have limited size which is a problem with Deep
Learning methods particularly.
This work aims to explore the impact of knowledge graphs’ semantics combined with the learning capabilities of graph neural networks to predict drug response. Specifically, how does using
semantics and considering graph data with context in patient drug response association prediction
and if knowledge-injected GNNs perform better than knowledge graph embedding approaches
To do this, two different approaches were developed: one using knowledge graph embeddings to
represent patients’ genetic information and their drug treatment, and another using a dual graph
neural network approach that works over the gene and drug knowledge graphs to learn patient and
drug representations for supervised learning.
The findings of this study were that the semantic context provided by knowledge graphs and
ontologies had a positive impact on patient-drug response prediction results. However, the GNNbased approach didn’t achieve better results than the knowledge graph embedding-based results,
despite only a slight difference in performance between them.
Descrição
Tese de mestrado, Engenharia Informática, 2025, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Previsão da resposta paciente-fármaco aprendizagem automática grafos de conhecimento redes neuronais Teses de mestrado - 2025