Evolving knowledge graph similarity for supervised learning in complex biomedical domains

Abstract

Background: In recent years, biomedical ontologies have become important for describing existing biological knowledge in the form of knowledge graphs. Data mining approaches that work with knowledge graphs have been proposed, but they are based on vector representations that do not capture the full underlying semantics. An alternative is to use machine learning approaches that explore semantic similarity. However, since ontologies can model multiple perspectives, semantic similarity computations for a given learning task need to be fine-tuned to account for this. Obtaining the best combination of semantic similarity aspects for each learning task is not trivial and typically depends on expert knowledge. Results: We have developed a novel approach, evoKGsim, that applies Genetic Programming over a set of semantic similarity features, each based on a semantic aspect of the data, to obtain the best combination for a given supervised learning task. The approach was evaluated on several benchmark datasets for protein-protein interaction prediction using the Gene Ontology as the knowledge graph to support semantic similarity, and it outperformed competing strategies, including manually selected combinations of semantic aspects emulating expert knowledge. evoKGsim was also able to learn species-agnostic models with different combinations of species for training and testing, effectively addressing the limitations of predicting protein-protein interactions for species with fewer known interactions. Conclusions: evoKGsim can overcome one of the limitations in knowledge graph-based semantic similarity applications: the need to expertly select which aspects should be taken into account for a given application. Applying this methodology to protein-protein interaction prediction proved successful, paving the way to broader applications.

A Collection of Benchmark Data Sets for Knowledge Graph-based Similarity in the Biomedical Domain

“…For instance, the PPI data sets also support prediction of PPIs based on semantic similarity, as done in Sousa et al . ( 48 ). Despite being domain-specific, we expect this collection to be useful beyond the biomedical domain.…”

“…Both Sousa et al . ( 48 ) and Maetschke et al . ( 4 ) show this to be true, whose PPI predicting approaches demonstrate higher predictive power when using terms from these two aspects.…”

“…The benchmark supports simple evaluation metrics, such as computing Pearson’s correlation coefficient, but it also supports more complex evaluations. For instance, the PPI data sets also support prediction of PPIs based on semantic similarity ( 48 ).…”

Supervised Semantic Similarity

“…Although there is a vast array of tunable parameters in every GP implementation, they usually do not substantially influence the quality of the evolved solutions [48]. For this reason, the parameters have the same values as proposed in [33]. For XGB, we use the XGBoost 1.1.1 package 7 , with the values of some parameters altered to maximize the performance of the model, through grid search.…”

“…In previous work, we developed a methodology to predict protein-protein interactions that uses genetic programming to evolve combinations of aspect-oriented semantic similarities [33]. The positive results inspired us to hypothesize that if data regarding a specific biomedical entity similarity is available (e.g, sequence similarity, domain similarity, etc), we can use it to effectively learn a supervised semantic similarity, that is able to combine different semantic aspects without user input, but is tailored to capture a specific biological similarity view.…”