建立最佳化的多基因風險分數預測台灣乳癌風險

Taipei Medical University Institutional Repository > 醫學科技學院 > 醫學資訊研究所 > 博碩士論文 > Item 987654321/63368

請使用永久網址來引用或連結此文件: http://libir.tmu.edu.tw/handle/987654321/63368

題名:	建立最佳化的多基因風險分數預測台灣乳癌風險 Developing an Optimized Polygenic Risk Score for Breast Cancer in Taiwan
作者:	王昱鈞 WANG, YU-CHUN
貢獻者:	醫學資訊研究所碩士班邵于宣蕭自宏
關鍵詞:	多基因風險分數;單核?酸多態性;全基因組關聯研究;乳癌風險預測 Polygenic risk score;Single nucleotide polymorphism;Genome-wide association study;Breast cancer risk prediction
日期:	2023-07-13
上傳時間:	2023-12-15 16:09:12 (UTC+8)
摘要:	研究動機：本研究旨在解決台灣乳癌風險預測中缺乏針對自己族群的多基因風險分數（Polygenic risk score, PRS）的問題，並希望建構出最佳化的PRS方法，以提升乳癌風險預測的效果。方法：使用台灣精準醫療計畫（Taiwan Precision Medicine Initiative, TPMI）的台中榮民總醫院參與者，共有1787名女性乳癌病患，並採用三種方式建構台灣人群的乳癌多基因風險分數。第一種方式當中利用歐洲預測效果最好的313個單核?酸多態性（Single Nucleotide Polymorphism, SNP）的PRS建構。第二種方式則是統整112個來自PGS catalog的乳癌PRS，並使用C＋T的方法作為建構。第三種方式中使用全基因組的資料作為輸入機器學習模型的特徵，透過Lasso regression進行特徵選擇。結果：利用歐洲預測效果最好的313個SNP的PRS建構出272-SNP PRS，在驗證集的效果為AUC＝0.580，OR per SD＝1.34（95％CI：1.12－1.59）。第二種方式統整112個來自PGS catalog的乳癌PRS，並使用C＋T的方法作為建構，其效果為AUC＝0.587，OR per SD=1.39（95％CI：1.17－1.66）。在第三種方式中使用全基因組的資料作為輸入機器學習模型的特徵，透過Lasso regression進行特徵選擇，選出742個SNP建構PRS，效果為AUC＝0.564，OR per SD＝1.27（95％CI：1.07－1.50）。結論：統整PGS catalog乳癌PRS而選擇出的SNP，在clumping R2參數為0.8，p-Value小於10-2的1,337個SNP，搭配272-SNP PRS有最佳的效果，AUC＝0.594，OR per SD＝1.42（95％CI：1.19－1.63）。為目前使用台灣族群發表的論文中預測效果最佳的。 Motivation: This study aims to address the lack of population-specific polygenic risk score (PRS) for breast cancer risk prediction in Taiwan and to enhance the effectiveness of breast cancer risk prediction through the development of optimized methods based on PRS. Methods: A total of 1787 female breast cancer patients from the Taiwan Precision Medicine Initiative (TPMI) at the Taichung Veterans General Hospital. The PRS for breast cancer in the Taiwanese population has been established using three distinct approaches. In the first approach, the PRS was constructed using the best-performing 313 single nucleotide polymorphisms (SNPs) for breast cancer risk prediction in European populations. In the second approach, 112 breast cancer PRS from the PGS catalog were integrated, and the clumping and p value thresholding (C+T) method was used for construction. The third approach involved using genetic variants data as input for a machine learning model, and Lasso regression was employed for feature selection. Results: a 272-SNP PRS for Taiwanese breast cancer was created using the best-performing 313-SNP PRS from European. Area under the curve (AUC) for the validation set was 0.58, and the odds ratio per standard deviation (OR per SD) was 1.34 (95% confidence interval [CI]: 1.12-1.59). In the second approach, 112 breast cancer PRS from the PGS catalog were integrated, and the performance was achieved using the C+T method. The AUC was 0.587, and the OR per SD was 1.39 (95% CI: 1.17-1.66). The third approach involved using genetic variants data as input for a machine learning model, and Lasso regression is used to select the features. The selected 742 SNPs were used to construct the PRS, and the results showed an AUC of 0.564 and an OR per SD of 1.27 (95% CI: 1.07 - 1.50). Conclusions: The SNP selection from integrating the PGS catalog for breast cancer PRS resulted in 1337 SNPs with a clumping R2 parameter of 0.8 and a p-Value less than 10-2. The highest performance was achieved by combining these 1337 SNPs with the previously developed 272-SNP PRS, which generated an AUC of 0.594 and an OR per SD of 1.42 (95% CI: 1.19-1.63). The prediction performance is optimal among the papers currently published using Taiwanese populations.
描述:	碩士指導教授:邵于宣共同指導教授:蕭自宏委員:張偉嶠委員:吳育瑋委員:李元綺委員:邵于宣委員:蕭自宏
資料類型:	thesis
顯示於類別:	[醫學資訊研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	98	檢視/開啟

在TMUIR中所有的資料項目都受到原著作權保護.

TAIR相關文章

資料載入中.....