GRU4ACE: Enhancing ACE inhibitory peptide prediction by integrating gated recurrent unit with multi-source feature embeddings
Issued Date
2025-06-01
Resource Type
ISSN
09618368
eISSN
1469896X
Scopus ID
2-s2.0-105005219361
Journal Title
Protein Science
Volume
34
Issue
6
Rights Holder(s)
SCOPUS
Bibliographic Citation
Protein Science Vol.34 No.6 (2025)
Suggested Citation
Ahmed S., Schaduangrat N., Chumnanpuen P., Shoombuatong W. GRU4ACE: Enhancing ACE inhibitory peptide prediction by integrating gated recurrent unit with multi-source feature embeddings. Protein Science Vol.34 No.6 (2025). doi:10.1002/pro.70026 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/110324
Title
GRU4ACE: Enhancing ACE inhibitory peptide prediction by integrating gated recurrent unit with multi-source feature embeddings
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Accurate identification of angiotensin-I-converting enzyme (ACE) inhibitory peptides is essential for understanding the primary factor regulating the renin-angiotensin system and guiding the development of new drug candidates. Given the inherent challenges in experimental processes, computational methods for in silico peptide identification can be invaluable for enabling high-throughput characterization of ACE inhibitory peptides. This study introduces GRU4ACE, an innovative deep learning framework based on multi-view information for identifying ACE inhibitory peptides. First, GRU4ACE utilizes multi-source feature encoding methods to capture the information embedded in ACE inhibitory peptides, including sequential information, graphical information, semantic information, and contextual information. Specifically, the feature representations used herein are derived from conventional feature descriptors, natural language processing (NLP)-based embeddings, and pre-trained protein language model (PLM)-based embeddings. Next, multiple feature embeddings were fused, and the elastic net was employed for feature optimization. Finally, the optimal feature subset with strong feature representation was input into a gated recurrent unit (GRU). The proposed GRU4ACE approach demonstrated superior performance over existing methods in terms of the independent test. To be specific, the balanced accuracy, sensitivity, and MCC scores of GRU4ACE reached 0.948, 0.934, and 0.895, which were 6.46%, 8.92%, and 12.51% higher than those of the compared methods, respectively. In addition, when comparing well-regarded feature descriptors, we found that the proposed multi-view features effectively captured crucial information, leading to improved ACE inhibitory peptide prediction performance. These comprehensive results highlight that GRU4ACE enhances prediction accuracy and significantly narrows down the search for new potential antihypertensive drugs.