Predicting protein-protein interactions in microbes associated with cardiovascular disease using deep denoising autoencoders and evolutionary information

Yang Li ^1* Senyu Zhou ² Jian Luo ² Mei Tang ² Chaojun Li ² Wenhua He ^2*

• ¹ Hefei University of Technology, Hefei, China
• ² The Fourth Hospital of Changsha (Integrated Traditional Chinese and Western Medicine Hospital of Changsha, Changsha Hospital of Hunan Normal University), Changsha, China

Protein-protein interactions (PPIs) are critical for understanding the molecular mechanisms underlying various biological processes, particularly in microbes associated with cardiovascular disease. Traditional experimental methods for detecting PPIs are often time-consuming and costly, leading to an urgent need for reliable computational approaches. In this study, we present a novel model, the deep Denoising Autoencoder for Protein-Protein Interaction (DAEPPI), which leverages the denoising autoencoder and CatBoost algorithm to predict PPIs from the evolutionary information of protein sequences. Our extensive experiments demonstrate the effectiveness of the DAEPPI model, achieving average prediction accuracies of 97.85% and 98.49% on yeast and human datasets, respectively. Comparative analyses with existing effective methods further validate the robustness and reliability of our model in predicting PPIs. Additionally, we explore the application of DAEPPI in the context of cardiovascular disease, showcasing its potential to uncover significant interactions that could contribute to understanding disease mechanisms. Our findings indicate that DAEPPI is a powerful tool for advancing research in proteomics and could play a pivotal role in the identification of novel therapeutic targets in cardiovascular disease.