基於表面電漿子共振技術與機器學習進行阿茲海默症表觀遺傳變化之模式辨識分析

Abstract

阿茲海默症（AD）為最常見之神經退化性疾病，主要影響高齡族群，臨床上以漸進性認知功能衰退為特徵。傳統病理機制聚焦於 β 類澱粉蛋白（Aβ）沉積與 tau 蛋白形成之神經纖維異常纏繞；然而，近年研究指出，表觀遺傳調控亦參與疾病早期發展，尤以 DNA 甲基化異常被視為具潛力之早期生物標記。本研究針對 HOXB6 基因的 DNA 甲基化位點進行分析，該位點於多項阿茲海默症患者的全基因組甲基化資料中呈現顯著過度甲基化，可能與神經發育基因的表現失衡相關，參與早期病理機制。為探討其作為生物標記之可行性，本研究設計模擬甲基化與非甲基化狀態之單股 DNA（ssDNA）探針，並建構基於表面電漿子共振（SPR）技術之光學偵測平台，以評估該序列於不同金表面修飾條件下的分子吸附動力學與光學訊號表現。研究中設計四種具不同親疏水性與官能基的金屬修飾表面，針對甲基化、未甲基化及控制組 ssDNA 分別進行 SPR 測定，建立訊號熱圖並進行可視化分析。為解決 SPR 底角辨識受光譜不對稱干擾之問題，進一步開發自適應底角搜尋演算法，採用四層決策樹（Decision Tree）架構以自動辨識 SPR 光譜之底角位置，平均誤差僅 2.8 mdeg，顯著優於多項式擬合與質心法等傳統方法。最終將訊號特徵與底角變化進行主成分分析（PCA）投影，結果顯示在單一最適表面條件下即可達成 97.5\% 的平均準確率，兩類樣本的 F1-score 均接近 0.97，產生具穩定性與高鑑別力之分群。整體而言，本研究成功結合表面修飾策略與光譜特徵分析，建立一套無標記 SPR 偵測平台，具潛力應用於阿茲海默症相關表觀遺傳變化之早期診斷。

Alzheimer's disease (AD) is the most common neurodegenerative disorder, primarily affecting the elderly population and clinically characterized by progressive cognitive decline. Traditionally, the pathological mechanisms of AD have been attributed to β-amyloid (Aβ) plaque accumulation and neurofibrillary tangles formed by tau protein. However, recent studies have indicated that epigenetic regulation also plays a critical role in the early development of the disease, with DNA methylation abnormalities recognized as potential early biomarkers. This study focuses on the DNA methylation site of the HOXB6 gene, which has been identified as significantly hypermethylated in multiple genome-wide methylation analyses of AD patient samples. Such abnormal methylation is presumed to be associated with impaired regulation of gene expression involved in neural development, potentially contributing to early pathogenic processes. To evaluate its biomarker potential, we designed synthetic single-stranded DNA (ssDNA) probes representing both methylated and unmethylated states of the target sequence, and established a label-free optical detection platform based on surface plasmon resonance (SPR) to assess the molecular absorbing dynamics and optical response under various gold surface modification conditions. Four gold surfaces with distinct hydrophobicity and functional group characteristics were modificated for the study. SPR measurements were conducted using methylated, unmethylated, and control ssDNA, followed by signal mapping and heatmap-based visualization. Furthermore, to address the issue of angle recognition errors caused by asymmetric SPR spectra, we developed an adaptive dip-angle detection algorithm based on a four-layer decision tree model, achieving an average prediction error of only 2.8 mdeg—markedly outperforming conventional polynomial fitting and centroid methods. Principal component analysis (PCA) was performed on the extracted signal features and dip-angle variations, revealing that under a single optimized surface condition, an average classification accuracy of 97.5\% was achieved. Both methylated and unmethylated samples yielded F1-scores of approximately 0.97, indicating highly stable and well-separated clustering performance. Overall, this study successfully combines surface modification strategies with spectral feature analysis to establish a label-free SPR detection platform with strong potential for the early diagnosis of Alzheimer's disease through epigenetic biomarker profiling.