應用跨研究之單核苷酸多態性標記子以建立整合性遺傳風險預測模型

Abstract

隨全基因組關聯性研究普及化，成百上千單核苷酸多態性標記子，與常見複雜疾病具有遺傳關聯性，然而關聯性顯著之單一標記子，顯少具備臨床應用等級之疾病判別力(勝算比>350)。遺傳風險預測研究(Genetic Risk Prediction Study)主要目的是將多重遺傳標記子重組累積風險量，以眾標記子之聯合判別力，重組高效能遺傳風險模型。然而標準建模流程是以世代研究或病例對照研究為實驗設計，進行全基因組關聯性研究，實驗成本昂貴，事後追蹤費時費力。本研究提出一套電腦模擬的演算流程，將跨研究之單核苷酸多態性標記子群集，整合成疾病風險評估模型與風險判別器，以節省建模成本。以五種常見複雜疾病，分別為心肌梗塞、銀屑病、大腸直腸癌、胃癌、一型糖尿病為範例，運用三種主要的概念，標記子標準化、資料探勘排除法、迭代性增加法，建立風險模型，有效地運用充足的標記子資源，以預測個人風險與疾病判別。可應用於個人生涯規劃、遺傳諮詢與疾病檢測平台設計，或設計世代研究及病例對照研究前之模型效能預估。

Genome-wide association study have surged in popularity in recent years, thousands of single nucleotide polymorphisms (SNPs) are associated with common complex disease potentially serve as biomarkers. However, needed sufficient odds ratios level (OR>350) are rarely observed when applying merely one SNP marker for discriminating between outcome and health people. In genetic risk prediction study, researchers cumulate genotype relative risks to one union risk and construct genetic risk prediction model with multiple markers to achieve adequate discriminative ability. Standard process for modeling is expensive and time-consuming such as cohort study or case-control experimental design. In this study, we develop an in silico algorithm integrating cross-study single nucleotide polymorphism data into genetic risk prediction models, and extend models to risk discriminators. Based on three major concepts, marker standardization, data mining and iteratively marker-adding to models, we efficiently reuse published SNPs to predict future health status and subclinical disease status for Asian population among selected five common complex diseases. We can apply the algorithm to individual life planning, genetic counseling, disease screening panel design, or prior estimation of a model performance.