摘要
目前基于建模的抗癌药物敏感性预测研究较多,但这些模型大多使用传统单任务学习模型。这种模型在解决复杂问题时需将问题拆分成单个子问题,忽略了各个子问题之间存在的关联,因而模型精度会受到影响。大多数药物敏感性预测模型仅使用了基因表达数据,忽略了基因突变、甲基化以及拷贝数等数据对药物敏感性预测的影响。结合上述数据,并考虑到不同药物之间可能存在的相似性,利用多任务学习方法共享任务之间的信息,对抗癌药物敏感性进行预测,预测的平均精度达到56%以上,较普通的Lasso模型提高了35%左右。同时,针对每种药物找出一些敏感的生物标志物,这些生物标志物可为癌症治疗提供指导。
Currently,in the study of cancer treatment,many experts and scholars are committed to the prediction of anti-cancer drug sensitivity based on modeling.However,most of the predictions for the sensitivity of anti-cancer drugs use the traditional single-task learning model.This kind of model needs to solve a complex problem by splitting the complex problem into a single sub-problem,and ignores the relationship between each sub-problem,so the accuracy of the model will be affected.At the same time,most drug sensitiv⁃ity prediction models only use gene expression data,and ignore the impact of gene mutation,methylation and copy number on drug sensitivity prediction.In the study,the above four kinds of data were combined,and considering the possible similarities between dif⁃ferent drugs,the multi-task learning method was used to share the information between tasks,and the anti-cancer drug sensitivity was predicted.The predicted average accuracy is over 56%,which is about 35%higher than the average Lasso model.At the same time,some sensitive biomarkers have been identified for each drug,which provide some guidance for the treatment of cancer.
作者
唐益翔
TANG Yi-xiang(School of Management,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《软件导刊》
2020年第1期207-210,共4页
Software Guide
关键词
癌症
药物敏感性
个性化医疗
多任务学习
预测
cancer
drug sensitivity
personalized medicine
multitask learning
prediction
