Development of Machine Learning-Based Models for Predicting Antimicrobial Peptides and Simulation Analysis of Membrane-Penetrating Properties

NCU Institutional Repository > 理學院 > 化學研究所 > 博碩士論文 > Item 987654321/97475

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/97475

题名:	Development of Machine Learning-Based Models for Predicting Antimicrobial Peptides and Simulation Analysis of Membrane-Penetrating Properties
作者:	廖柏鈞;Liao, Bo-Jyun
贡献者:	化學學系
关键词:	機器學習;抗菌肽;Machine Learning;Antimicrobial Peptides
日期:	2025-07-29
上传时间:	2025-10-17 11:25:10 (UTC+8)
出版者:	國立中央大學
摘要:	陽離子短肽（如 TAT）因其細胞穿透與藥物遞送潛力，成為生物醫藥領域的研究焦點。然而，TAT 與細胞膜交互作用的機制仍不明確，尤其是在不同插入深度與多肽數量下對脂雙層結構的影響。本研究結合全原子與粗粒化分子動力學模擬，系統探討 TAT 與膜的交互作用及其作用機制。全原子模擬顯示，TAT 在淺層插入時會引發磷脂質頭基的重新排列、局部膜厚增加及有序性下降，並且穿膜自由能隨著TAT數量增加而下降。在粗粒化模擬中，我們建立了包含 2876 條脂質與 575 條 TAT 的大型囊泡模型，TAT 的大量吸附會降低囊泡圓度並造成局部凹陷，進一步顯示 TAT 透過在膜面聚集來誘發局部曲率改變，而非傳統的穿孔機制。這些結果顯示多尺度模擬在研究膜動態行為中的潛力。此外，我們根據世界衛生組織細菌優先病原體清單，開發了多個機器學習模型來預測潛在的抗菌肽（AMPs）。這些模型旨在大量且有效率地篩選肽分子，快速篩選出有潛力成為AMPs的肽分子，為後續的分子動力學模擬和實驗驗證提供優先選擇。儘管大部分針對不同細菌的模型在測試上表現良好，但少數不好的模型表現也突顯了調整篩選標準的重要性，未來研究將專注於優化特徵選擇以提高預測可靠性，並進一步支持新型抗菌肽的設計與篩選。;Cationic short peptides, such as TAT, have become a research focus in the biomedical field due to their potential for cell penetration and drug delivery. However, the mechanism of TAT’s interaction with cell membranes is still unclear, especially regarding its impact on lipid bilayer structure at different insertion depths and peptide concentrations. In this study, we used both all-atom and coarse-grained molecular dynamics simulations to systematically investigate the interaction between TAT and membranes, as well as its possible mechanisms of action. All-atom simulations showed that when TAT inserts shallowly into the membrane, it causes rearrangement of the lipid headgroups, an increase in local membrane thickness, and a decrease in lipid tail ordering. The membrane insertion free energy decreases as the number of TAT peptides increases. In the coarse-grained simulations, we constructed a large vesicle model consisting of 2876 lipids and 575 TAT peptides. The large adsorption of TAT peptides reduced the vesicle′s roundness and caused local indentation, further indicating that TAT induces changes in local membrane curvature through accumulation on the membrane surface, rather than the traditional pore formation mechanism. These results demonstrate the potential of multiscale simulations in studying membrane dynamics. Additionally, based on the WHO priority pathogens list, we developed multiple machine learning models to predict potential antimicrobial peptides (AMPs). These models are designed to quickly and efficiently screen large numbers of peptide molecules, providing prioritized candidates for subsequent molecular dynamics simulations and experimental validation. Although most models performed well in testing for different bacterial species, the few models with poor performance also highlighted the importance of adjusting the selection criteria. Future research will focus on optimizing feature selection to improve prediction reliability and further support the design and screening of novel antimicrobial peptides.
显示于类别:	[化學研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	27	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....