Gene Function Annotation in the Renin-Angiotensin System (RAS) Using Multi-Label Deep Learning

The renin-angiotensin system (RAS) plays a crucial role in regulating blood pressure and fluid homeostasis. Angiotensin-II (Ang-II) exerts its effects through the Ang-II type 1 receptor (AT1R), leading to vasoconstriction, sodium reabsorption, and increased blood pressure. This study aims to employ deep learning and text mining methods to enhance gene function annotation in the RAS. A multi-label gene function annotation model was designed based on a deep learning strategy, for gene multi-function discovery of the RAS and its relevance to pathways using full-text biomedical publications. A deep learning algorithm was trained based on the Backpropagation Multi-Label Learning method, to predict gene multi-function using the collected data matrix. A data matrix consisting of1026 genes names and 39463 full-text publications related to these genes, was collected from Pubmed and PMC. To decrease the data dimension of the problem as much as possible to shorten the training time of the learning model, we applied a deep-learning-based model, Stacked Denoising Autoencoder Multi-Label Learning (SdaMLL), for facilitating gene multifunction discovery and pathway completion. Experimental results indicate that SdaMLL outperforms existing classical multi-label algorithms. Moreover, we found that some genes, like LNPEP (also known as IRAP) may exhibit a dual role within the RAS. We identified that LNPEP interacts strongly with proteins YWHAH and YWHAQ (Tyrosine 3-Monooxygenase/Tryptophan 5-Monooxygenase Activation Proteins). Through their encoded 14-3-3 proteins, these interactions are hypothesized to modulate the RAS by influencing downstream pathways of AT1R, eNOS, aldosterone synthesis, and renin release. These findings underscore the potential of 14-3-3 proteins as key modulators of cardiovascular function and blood pressure regulation. By addressing traditional methods' limitations and leveraging advanced computational intelligence and machine learning methods, we provided a comprehensive understanding of the RAS and its pathways. This will facilitate further research in cardiovascular biology and set a new model for applying deep learning in biomedical research, paving the way for future innovations and discoveries. This work was supported in part by a research grant from the National Institutes of Health (HL163588). This abstract was presented at the American Physiology Summit 2025 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.