基於生物醫學文獻建構路徑重要性模型預測藥物新適應症

Abstract

藥物開發成本高昂且費時，根據美國食品藥品管理局(FDA)規定，新藥物需通過候選藥物開發、臨床實驗、FDA審核等五大流程，最終才可以在市場中販售。然而，只要其中一個流程無法通過，此藥物開發就前功盡棄，投資者將承受巨大損失。因此，為了解決藥物開發的困難，許多研究人員開始尋求替代方法。「舊藥新用」透過既有藥物，尋找新適應症，能夠大幅降低藥物開發金錢、時間成本。 Swanson (1986)最先提出以醫學文獻探勘方式實現舊藥新用，然而，之後依據Swanson模型的研究碰到許多困難。因此，我們提出基於生物醫學文獻建構路徑重要性模型以預測藥物新適應症的方法。首先我們會以醫療語意關係建立語意網路，接著建置分類模型學習區分路徑重要性，最後依照分類模型結果對候選疾病進行排序，找出最有可能的藥物新適應症。 我們以實驗證明我們提出的路徑重要性分類模型有不錯的表現水準，並證明與傳統方法相比，融入區分路徑重要性模型能夠更有效找出潛在藥物新適應。

Drug development is costly and time-consuming. According to United States Food and Drug Administration (FDA), drug development consists of five stages, including drug discovery, clinical test, FDA review, etc. However, once one of the stages fails, the investment on candidate drug seldom returns. As a result, to overcome the challenges of drug development, researchers start to explore alternative methods for drug development. Drug repurposing discovery, finding new indications for existing drugs, has been proposed to help reduce cost and time needed for drug development. Swanson (1986) originally proposed a drug repurposing approach that analyzes biomedical literatures to uncover implicit relationships. Previous studies following Swanson’s ABC model encountered several limitations. Therefore, in this research, we propose a path-importance-based approach, which constructs a concept network based on semantic predication, trains a classification model to determine the importance of paths that connecting a focal drug and a candidate disease, and finally ranks candidate diseases according to the importance of paths identified by the path importance classification model. In our systematic evaluation experiments, we prove that our path importance classification model achieves a satisfactory effectiveness, and that adopting the concept of path importance into the ranking of candidate drugs for drug repurposing outperforms the traditional method.