結合臨床與個人基因特徵預測24個月後之簡易心智量表（MMSE）變化

Abstract

近年來醫療發展快速，衛生環境提升，平均壽命明顯的提升，隨著平均年齡的上升，老年退化顯然已經成為造成高齡化社會一定會面臨到的大問題。其中，阿茲海默症（Alzheimer's disease）是俗稱的老年失智症中最常見的疾病類型，阿茲海默氏症最早於1906年由德國精神病學家和病理學家愛羅斯•阿茲海默首次發現，已經發現超過一世紀之久，但到今天科學家們依舊沒有可以阻止或逆轉病程的治療，只有少數可能可以暫時改善症狀的方法。 阿茲海默症的病人將會漸漸失去他們的記憶和他們的認知能力，甚至是性格也都會產生很大的變化，有時會伴隨著憂鬱、偏執和妄想症等，其主要原因為處理訊息儲存的神經細胞凋亡。目前，全球身受阿茲海默症影響的病人約2400∼3500萬人，加上人口平均年齡上升，預計到了2050年，每85個人就會有一個人有阿茲海默症。我們必須正視這個問題，了解阿茲海默症形成的原因從根本解決這個疾病。 阿茲海默症被認為是一種由於基因的變異造成的複雜遺傳疾病。晚發性的阿茲海默症病人患病率高但病人個體間的基因變異複雜，這樣的現象造成疾病判斷的不易。根據以往的研究，APOEε-4及其等位基因，為目前為止唯一確信的阿茲海默症遺傳因子；除此之外，還有數以百計的高風險基因變異被認為與阿茲海默症有高度相關，但其關係尚不清楚。 本研究收集ADNI資料庫中病人的臨床及基因特徵，分析基因變異及簡易心智量表（MMSE量表）的變化下手，利用全基因組關聯分析（Genome-wide association study, GWAS）從約一百五十萬個單核苷酸多態性（SNP）中，找出了39個和智能退化高度相關的單核苷酸多態性，同時將提供一個有效的解決方案來選擇臨床樣本資料，進行臨床試驗和早期疾病治療的可能性。本研究並利用機器學習演算法支持向量機（Support Vector Machines）建構出預測阿茲海默症退化進程的模型，在訓練資量上，皮爾遜相關係數可達到0.5，在預測兩份獨立的資料上，皮爾遜相關係數為0.43與0.35。本論文在做全基因體關聯分析時，使用不同的篩選條件，篩選出866個與智能退化相關的單核苷酸多態性，分佈在120個基因上。 此研究提供了一個有效的解決方案來篩選基因特徵與臨床樣本資料，建構智能退化進程模型，並將所篩選出的基因清單與目前已知和阿茲海默症的相關基因相互對照，期能尋獲與進程發展較直接相關的基因指標，來增加早期治療疾病的可能性。

The dramatic rise in life expectancy in the past few decades has resulted in a huge number of individuals achieving the age at which neurodegenerative disorders become common. Alzheimer's disease (AD) is one of the most common neurodegenerative disease discovered more than one century ago and also one of the most common elderly diseases in the world. Slowly but surely, AD patients will lose their memory and their cognitive abilities, and even their personalities may change dramatically. These changes are due to the progressive dysfunction and death of nerve cells that are responsible for the storage and processing of information. Currently, AD affects about 24 to 35 million people around the world. Combined with an aging population, prevalence is expected to increase to 1 in 85 people by 2050. In order to deal with the massive growth of the AD patients, it is important to find the mechanism of Alzheimer’s disease development. Alzheimer’s disease is known as a genetically complex and heterogeneous disorder disease. The late-onset Alzheimer’s disease is modulated by genetic variants with relatively low penetrance but high prevalence. Based on previous studies, the only firmly established genetic susceptibility factor for Alzheimer’s disease is the ε-4 allele of APOE. Beyond this, hundreds of other putative risk alleles in other genes were reported. But the relationships between these published alleles and the Alzheimer’s disease still remain unclear. Both of the clinical and genetic data we used in this study were provided by the Alzheimer's Disease Neuroimaging Initiative (ADNI). To tackle the complex genetic variations of AD, this study aims to link not only the genetic (Genome-wide association study, GWAS) but also clinical data to the change of the 24 month follow-up cognitive scores (measured in the end of the 24-th month after initial assessment) by the machine learning algorithm, SVR (Support Vector Machine Regression, SVR). We retrieved 39 SNPs (Single Nucleotide Polymorphism, SNP) from 1.5 million SNPs that were shown to be highly correlated to the degeneration of Alzheimer’s disease. We built the predictive model using both clinical and genetic data, and the resultant Pearson correlation coefficient between the measured and the predicted scores is about 0.5 on one training data set and are 0.43 and 0.35 on two independent test data sets. With a relaxed threshold, we extracted 866 SNPs from 1.5 million SNPs in 120 genes that were shown to be highly correlated to the degeneration of Alzheimer’s disease. The constructed model not only can help to predict cognitive trajectory and provide new approaches for early identification of AD, but also provide an efficient solution to select the samples for clinical trials for earlier disease treatment.