Guiding D-amino Acid Containing Antimicrobial Peptide Design by Protein Structural Ensemble Informed Graph Deep Learning

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/97478

Title:	Guiding D-amino Acid Containing Antimicrobial Peptide Design by Protein Structural Ensemble Informed Graph Deep Learning
Authors:	吳紹麒;Wu, Shao-Chi
Contributors:	化學學系
Keywords:	抗菌肽;深度學習;圖神經網路;Antimicrobial Peptide;Graph Deep Learning
Date:	2025-07-29
Issue Date:	2025-10-17 11:25:31 (UTC+8)
Publisher:	國立中央大學
Abstract:	抗生素的廣泛使用促使了多重抗藥性細菌的出現，這些細菌引發的感染無法透過現有的抗生素進行有效治療，因此抗微生物肽（AMPs）被視為重要的替代治療方案。隨著人工智慧的快速發展，已經有許多高準確度的機器學習模型能夠加速AMPs的開發。然而，儘管D-胺基酸的引入是一種常見的用於提升肽類穩定性的設計策略，現有的機器學習模型大多未考慮D-胺基酸在AMPs中的存在。因此，開發一個能夠評估含有D-胺基酸的AMPs抗菌活性的模型顯得尤為迫切。本研究嘗試通過有向消息傳遞神經網絡（D-MPNN）將含有D-胺基酸的AMPs的三維結構以圖的形式引入深度學習模型中。為了獲得這些胜肽的結構，我們結合AlphaFold2和分子動力學模擬從序列中預測其結構。根據不同的結構預測方法，我們發現D-MPNN能夠將AMP活性與胜肽的結構系集（structural ensemble）更好地關聯，在獨立測試集上達到了82.24%的準確度。此外，我們還進一步探討了該模型的應用性。常用的胜肽設計策略包括單一殘基D-胺基酸取代、全序列D-胺基酸取代以及逆向倒位（retro-inverso），本模型能夠被用於預先評估這些策略的可行性；同時，透過夏普力值（Shapley value）分析，該模型亦具有改進低效價AMPs的潛力。;The widespread use of antibiotics has led to the emergence of multi-drug-resistant bacteria, which cause infections that cannot be effectively treated with existing antibiotics. As a result, antimicrobial peptides (AMPs) are regarded as important alternative therapeutics. With the rapid development of artificial intelligence, numerous high-accuracy machine learning models have been established to accelerate the discovery of AMPs. However, most existing machine learning models do not account for AMPs containing D-amino acids, despite the fact that D-amino acids incorporation is a common design strategy for improving peptide stability. Therefore, there is an urgent need to develop a model capable of evaluating the antimicrobial activity of D-amino acid containing AMPs. This study attempts to incorporate the 3D structures of D-amino acid containing AMPs into a deep learning framework using a Directed Message Passing Neural Network (D-MPNN). To obtain the structures of these peptides, we employed AlphaFold2 combined with molecular dynamics simulations to predict the 3D structures based on peptide sequences. Depending on the structural prediction method, we found that D-MPNN was able to better correlate AMP activity with the structural ensemble of peptides, achieving 82.24% accuracy on an independent test dataset. Furthermore, we explored the practical application of the developed model. Common peptide design strategies, including single-residue D-amino acid substitution, all D-amino acid substitution, and retro-inverso transformations, can be pre-assessed for their feasibility using this model. With the aid of Shapley value analysis, the model also has the potential to optimize low-potency AMPs, enhancing their effectiveness for future applications.
Appears in Collections:	[Graduate Institute of Chemistry] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	44	View/Open

社群 sharing

Loading...