Kuang Zhaobin, Thomson James, Caldwell Michael, Peissig Peggy, Stewart Ron, Page David
University of Wisconsin-Madison.
Morgridge Institute for Research.
IJCAI (U S). 2016 Jul;2016:2521-2528.
Computational Drug Repositioning (CDR) is the knowledge discovery process of finding new indications for existing drugs leveraging heterogeneous drug-related data. Longitudinal observational data such as Electronic Health Records (EHRs) have become an emerging data source for CDR. To address the high-dimensional, irregular, subject and time-heterogeneous nature of EHRs, we propose Baseline Regularization (BR) and a variant that extend the one-way fixed effect model, which is a standard approach to analyze small-scale longitudinal data. For evaluation, we use the proposed methods to search for drugs that can lower Fasting Blood Glucose (FBG) level in the Marshfield Clinic EHR. Experimental results suggest that the proposed methods are capable of rediscovering drugs that can lower FBG level as well as identifying some potential blood sugar lowering drugs in the literature.
计算药物重新定位（CDR）是利用异构药物相关数据为现有药物寻找新适应症的知识发现过程。纵向观察数据，如电子健康记录（EHR），已成为CDR的新兴数据源。为了解决EHR的高维、不规则、个体和时间异质性问题，我们提出了基线正则化（BR）及其扩展单向固定效应模型的变体，单向固定效应模型是分析小规模纵向数据的标准方法。为了进行评估，我们使用所提出的方法在马什菲尔德诊所电子健康记录中寻找能够降低空腹血糖（FBG）水平的药物。实验结果表明，所提出的方法能够重新发现可降低FBG水平的药物，并在文献中识别出一些潜在的降血糖药物。
