BioPrediction-PPI: Simplifying the Prediction of Protein-Protein Interactions through Artificial Intelligence

Abstract

Proteins are essential in biological processes, primarily through their interactions with other molecules, including proteins. These interactions are crucial for cellular functions and maintaining life. Predicting Protein-Protein Interactions (PPIs) is very important, although challenging, for understanding cellular functions and diseases. This paper presents BioPrediction-PPI, a new end-to-end Machine Learning (ML) framework for PPI prediction, automating the entire process from feature extraction to interpretability, with no manual intervention required. BioPrediction-PPI stands as one of the few end-to-end models that do not rely on deep learning, enabling the automated use of trained models on new data. It offers interpretability graphs for creating auditable models and has been evaluated through comparative experiments with 30 previous studies, using 15 datasets. Additionally, the interpretability graphs offer valuable insights for model evaluation and experimental design, facilitating informed decision-making. BioPrediction-PPI demonstrates competitive performance across multiple datasets, even without the use of deep learning, and is a transparent, white-box model that can be easily used by biologists and practitioners without a background in computer science. The proposed framework has the potential to accelerate research in biology and related fields by making AI-driven tools more accessible.