摘要
药物风险-效益评价中的一个重要问题是确定药物副作用的频率.相较于通常的随机对照实验,基于机器学习预测药物副作用频率的方法具有时间短、准确率高的特点,并且可以用来指导对照实验.现有的计算方法很少考虑“相似的药物具有相似的副作用频率”这一特点,因此预测性能仍有待进一步提高.本文提出结合拉普拉斯正则化的非负矩阵分解方法,并引入超参数控制未知副作用标签及其预测值的间隔.计算实验表明,该方法可以有效预测药物的副作用频率,并且还可以预测上市后药物的副作用.
An important issue in drug risk-benefit assessment is to determine the frequency of drug side effects.Compared with the usual randomized controlled trials,the method based on machine learning to predict the frequency of drug side effects has the characteristics of short time and high accuracy,and can be used to guide controlled trials.However,existing computational methods rarely take into account the feature that“similar drugs have similar frequency of side effects”,so the prediction performance can be further improved.Therefore a non-negative matrix factorization method combined with Laplace regularization is proposed in this article,and a hyperparameter is also introduced to control the margin between labels and their predicted scores for unknown side effects.Computational experiments show that this method can effectively predict the frequency of drug side effects,and can also predict post-marketing drug side effects.
作者
王林
李冰纯
徐显嵛
WANG Lin;LI Bingchun;XU Xianyu(College of Artificial Intelligence,Tianjin University of Science&Technology,Tianjin 300457,China)
出处
《天津科技大学学报》
CAS
2022年第3期67-72,共6页
Journal of Tianjin University of Science & Technology
基金
天津市教委科研计划项目(2018KJ107)。
关键词
药物
副作用频率
机器学习
拉普拉斯正则化
drug
side effect frequency
machine learning
Laplace regularization
