LASIGE researchers Rita Sousa, Sara Silva, and Catia Pesquita have published a study in Computers for Biology and Medicine, a top 10% journal, introducing KGsim2vec, a method that enhances the explainability of AI models for protein–protein interaction prediction. While machine learning is revolutionizing biomedical research, many models function as opaque “black boxes,” making it challenging to understand how they generate predictions. KGsim2vec overcomes this by leveraging knowledge graph-based semantic similarity, providing more transparent and interpretable AI-driven insights. The study shows that KGsim2vec maintains strong predictive accuracy while uncovering biological patterns and data biases, marking a significant step toward more explainable AI in life sciences.

The paper is available here.