AI-Augmented Bioinformatics Framework for Predicting Protein–Protein Interactions in Complex Diseases

Abstract

Protein-protein interactions (PPIs) are crucial to cellular regulation and they are involved in the pathogenesis of complex diseases but have been slow to be experimentally detected due to their high cost and low throughput. Recently, artificial intelligence (AI) has become a powerful tool to speed up the PPI forecasting process by integrating sequence, structural and multi-omics data. Based on this promise, we unveil a graph neural network-based, transformer-based sequence encodings-based, disease-specific knowledge graph-based bioinformatics framework that models PPIs of such conditions as cancer, Alzheimer disease, and autoimmune disorders. The framework combines a variety of omics data with structural bioinformatics piping with a prediction accuracy of 92% on benchmark datasets marking an improvement of 12-15% above standard machine learning strategies. The extension to case studies of breast cancer and Alzheimer disease further illustrates that the framework can be used in identifying new interactions that are disease-related, opening new possibilities of having therapeutic targets. These discoveries underline the usefulness of AI-added bioinformatics in the future evolution of computational prediction to translational medicine.