Enhanced Alzheimer’s Disease Prediction Through Integration of Protein-Protein Interaction Data and Meta-Learning

Association between proteins and diseases have been widely studied to understand disease triggers and identify potential therapeutic targets. Deciphering the complexity of gene networks is crucial for understanding diseases. Node embedding offers a powerful approach, revealing latent patterns that can predict novel gene-disease associations. In this study, we applied a meta-learning approach to predict protein-disease associations with a focus on Alzheimer’s disease. Our approach utilized curated data from the Disease Gene Network Database (DisGeNET), a database of gene-disease associations, and protein-protein interaction data from the Search Tool for Retrieval of Interacting Genes/Proteins (STRING) dataset. We generated embeddings for Protein-Protein Interactions(PPI) using the node2vec algorithm, capturing interaction patterns in a low-dimensional space. HybridBoost, a two-layered model architecture, was then employed, with random forest and CatBoost classifiers in the first layer and CatBoost as the meta-learner in the second layer. Our model achieved an accuracy of 96% and with precision, recall, and F1 scores also outperforming individual models like random forest, support vector machine, AdaBoost, and CatBoost. These results demonstrate the effectiveness of the proposed meta-learning model in predicting protein-disease associations for Alzheimer’s disease.