Multi-AOP: a lightweight multi-view deep learning framework for antioxidant peptide discovery

Abstract

Antioxidant peptides (AOPs), with their strong free radical scavenging ability and health benefits, have emerged as promising candidates for disease prevention and food preservation. However, traditional experimental approaches to AOP discovery remain hindered by inefficiencies and substantial resource demands. Here, we present Multi-AOP, a parameter lightweight multi-view deep learning framework (0.75 million parameters) that enhances AOP discovery through integrated sequence and graph learning. We employ Extended Long Short-Term Memory (xLSTM) to generate sequence embeddings. Concurrently, we transform peptide sequences into SMILES representations and extract molecular graph features using a Message Passing Neural Network (MPNN), capturing intrinsic physicochemical properties. By leveraging both sequence patterns and structural information through hierarchical fusion, Multi-AOP achieves accuracies of 0.8043, 0.9684, and 0.9043 on the AnOxPePred, AnOxPP, and AOPP benchmark datasets, respectively, consistently outperforming conventional machine learning algorithms and state-of-the-art deep learning approaches. Furthermore, we constructed a unified AOP dataset by integrating these benchmark datasets, facilitating the future development of generalizable AOP models. All datasets and the optimized predictive model are publicly accessible at https://github.com/CaiJianxiu/Multi-AOP.

References

[1] Apel K, Hirt H. Reactive oxygen species: metabolism, oxidative stress, and signal transduction. Annu Rev Plant Biol, 2004, 55(1): 373-399

[2] Badrinarayanan S, Guntuboina C, Mollaei P, Barati Farimani A. Multi-peptide: multimodality leveraged language-graph learning of peptide properties. Journal of Chemical Information and Modeling, 2024, 65(1): 83-91

[3] Beck M, Pöppel K, Spanring M, Auer A, Prudnikova O, Kopp M, Klambauer G, Brandstetter J, Hochreiter S (2024) xlstm: Extended long short-term memory. arXiv preprint arXiv:2405.04517

[4] Bryant P, Pozzati G, Elofsson A. Improved prediction of protein-protein interactions using alphafold2. Nat Commun, 2022, 13(1): 1265

[5] Cai J, Yan J, Un C, Wang Y, Campbell-Valois FX, Siu SWI. Bert-ampep60: A bert-based transfer learning approach to predict the minimum inhibitory concentrations of antimicrobial peptides for escherichia coli and staphylococcus aureus. Journal of Chemical Information and Modeling, 2025, 6573186-3202

[6] Choe E, Min DB (2005) Chemistry and reactions of reactive oxygen species in foods. Journal of food science, 70(9):R142–R159

[7] Dauparas J, Anishchenko I, Bennett N, Bai H, Ragotte RJ, Milles LF, Wicky BIM, Courbet A, de Haas RJ, Bethel Net al.. Robust deep learning-based protein sequence design using proteinmpnn. Science, 2022, 378(6615): 49-56

[8] Elias RJ, Kellerby SS, Decker EA. Antioxidant activity of proteins and peptides. Critical reviews in food science and nutrition, 2008, 48(5): 430-441

[9] Elnaggar A, Heinzinger M, Dallago C, Ghalia Rehawi Yu, Wang LJ, Gibbs T, Feher T, Angerer C, Steinegger Met al.. Prottrans: Toward understanding the language of life through self-supervised learning. IEEE Trans Pattern Anal Mach Intell, 2021, 44(10): 7112-7127

[10] Gilmer J, Schoenholz SS, Riley PF, Vinyals O, Dahl GE (2020) Message passing neural networks. In Machine learning meets quantum physics, 199–214. Springer,

[11] Graves A, Graves A (2012) Long short-term memory. Supervised sequence labelling with recurrent neural networks, 37–45

[12] Hamilton W, Ying Z, Leskovec J (2017) Inductive representation learning on large graphs. Advances in neural information processing systems, 30

[13] Irina Georgiana Munteanu and Constantin Apetrei (2021) Analytical methods used in determining antioxidant activity: A review. Int J Mol Sci 22(7):3380

[14] Jomova K, Raptova R, Alomar SY, Alwasel SH, Nepovimova E, Kuca K, Valko M. Reactive oxygen species, toxicity, oxidative stress, and antioxidants: Chronic diseases and aging. Archives of toxicology, 2023, 97(10): 2499-2574

[15] Khezerlou A, pouya Akhlaghi A, Alizadeh AM, Dehghan P, Maleki P, (2022) Alarming impact of the excessive use of tert-butylhydroquinone in food products: A narrative review. Toxicology reports 9:1066–1075

[16] Landrum G. Rdkit documentation Release, 2013, 1(1–79): 4

[17] Li J, Yang Y, Chen R, Zheng D, Pang PCI, Lam CK, Wang Y. Identifying healthcare needs with patient experience reviews using chatgpt. PLoS One, 2025, 203 e0313442

[18] Lian-Jiu S, Zhang J-H, Gomez H, Murugan R, Hong X, Dongxue X, Jiang F, Peng Z-Y. Reactive oxygen species-induced lipid peroxidation in apoptosis, autophagy, and ferroptosis. Oxid Med Cell Longev, 2019, 201915080843

[19] Li W, Liu X, Liu Y, Zheng Z (2025) High-accuracy identification and structure–activity analysis of antioxidant peptides via deep learning and quantum chemistry. Journal of Chemical Information and Modeling,

[20] Lin Z, Akin H, Rao R, Hie B, Zhu Z, Lu W, dos Santos Costa A, Fazel-Zarandi M, Sercu T, Candido S, et al (2022) Language models of protein sequences at the scale of evolution enable accurate structure prediction. BioRxiv, 2022:500902

[21] Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Advances in neural information processing systems, 30

[22] Olsen TH, Yesiltas B, Marin FI, Pertseva M, García-Moreno PJ, Gregersen S, Overgaard MT, Jacobsen C, Lund O, Hansen EBet al.. Anoxpepred: using deep learning for the prediction of antioxidative properties of peptides. Scientific reports, 2020, 10121471

[23] Othman A, Norton L, Finny AS, Andreescu S. Easy-to-use and inexpensive sensors for assessing the quality and traceability of cosmetic antioxidants. Talanta, 2020, 208 120473

[24] Pop A, Kiss B, Loghin F. Endocrine disrupting effects of butylated hydroxyanisole (bha-e320). Clujul Medical, 2013, 86116

[25] Qin D, Jiao L, Wang R, Zhao Y, Hao Y, Liang G. Prediction of antioxidant peptides using a quantitative structure- activity relationship predictor (anoxpp) based on bidirectional long short-term memory neural network and interpretable amino acid descriptors. Comput Biol Med, 2023, 154 106591

[26] Saravanakumar G, Kim J, Kim WJ. Reactive-oxygen-species-responsive drug delivery systems: promises and challenges. Advanced Science, 2017, 411600124

[27] Sumida KH, Núñez-Franco R, Kalvet I, Pellock SJ, Wicky BIM, Milles LF, Dauparas J, Wang J, Kipnis Y, Jameson Net al.. Improving protein expression, stability, and function with proteinmpnn. Journal of the American Chemical Society, 2024, 14632054-2061

[28] Tang S, Li B, Yu H (2024) Chebnet: efficient and stable constructions of deep neural networks with rectified power units via chebyshev approximation. Communications in Mathematics and Statistics, 1–27

[29] Tao H, Wang X, Huang S-Y (2025) An interaction-derived graph learning framework for scoring protein–peptide complexes. Nature Machine Intelligence, 1–12

[30] Veličković P, Cucurull G, Casanova A, Romero A, Lio P, Bengio Y (2017) Graph attention networks. arXiv preprint arXiv:1710.10903,

[31] Xu K, Hu W, Leskovec J, Jegelka S (2018) How powerful are graph neural networks? arXiv preprint arXiv:1810.00826

[32] Yan K, Lv H, Guo Y, Peng W, Liu B (2023) samppred-gat: prediction of antimicrobial peptide by graph attention network and predicted peptide structure. Bioinformatics, 39(1):btac715

[33] Zhenjiao D, Caragea D, Guo X. and Yonghui Li, 2025, Pepbert, Lightweight language models for bioactive peptide representation. bioRxiv

Funding

Macao Polytechnic University(RP/FCA-06/2024)

the Science and Technology Development Fund from Macao S.A.R.(0021/2024/AMR)

RIGHTS & PERMISSIONS

The Author(s)