基于Stacking算法与多数据源的有效药物组合

Effective Drug Combination Based on Stacking Algorithm and Multi-data Source

对于癌症、心血管疾病等复杂疾病,采取组合用药克服耐药性和改善功效已成为标准治疗方案。鉴定药物组合标准的方法是进行体内或体外药物筛选实验,但这一过程很缓慢,代价高昂。各种高通量组学技术产生度量药物效应的各层次数据,使得从计算角度挖掘数据进而预测有效药物组合成为主流手段。针对有效药物组合的预测模型大多是利用单一机器学习模型建模。为获得更高的精度,提出一种新的有效药物组合预测方法。该方法充分利用5种不同层次的药物信息构建相似性特征,特别引入药物靶标的序列信息和功能信息,基于Stacking算法融合多个传统机器学习模型和最新的集成学习模型LightGBM。实验表明,该方法预测的AUC值为0.953,精度比单一机器学习模型有显著提升。

For complex diseases such as cancer and cardiovascular diseases,it has become a standard treatment to use combination drugs to overcome drug resistance and improve efficacy.The method for identifying drug combination criteria is through in vivo or in vitro drug screening experiments,but this process is expensive and slow.Various high-throughput omics techniques produce data at various levels that measure drug effects,making it a mainstream and effective means of mining data from a computational perspective to predict effective drug combinations.Most of the current predictive models for effective drug combinations are modeled using a single machine learning model.In order to obtain a better prediction rate,this paper proposes a new effective drug combination prediction method.The method makes full use of five different levels of drug information to construct similarity features,especially the introduction of sequence information and functional information of drug targets.Based on the Stacking algorithm,multiple traditional machine learning models and the latest integrated learning model LightGBM are combined.Finally,the AUC value of the prediction result of this method is as high as 0.953,and the precision is significantly improved compared with the single machine learning model.