癌症的耐药性预测任务已经成为精准医学领域前瞻性研究方向之一。针对现有耐药性预测方法难以深度表征药物和细胞系之间协同关系的问题,提出一种非线性子空间驱动下的耐药性预测方法NLS-DRP(Nonlinear Subspace-Driven Drug Resistance Prediction)。NLS-DRP包括Cell分支、Drug分支和协同融合三个关键学习模块,分别用于构建非线性子空间提取细胞系特征,拆分药物结构提取子序列特征,设计非线性协同空间融合细胞系和药物特征;最后,通过融合三个模块的特征,实现细胞系对药物的耐药性预测。在癌症药物敏感性基因组学和癌症细胞系百科全书两个公开数据集上进行实验,结果表明所提NLS-DRP模型显著优于对比的基准方法,取得了0。945 8的皮尔逊相关系数(PCC)值和0。924 2的斯皮尔曼相关系数(SCC)值,验证了本文方法的有效性。
Nonlinear Subspace-Driven Drug Resistance Prediction
The task of predicting drug resistance in cancer has emerged as a prospective research direction in the field of precision medicine.To address the challenge of limited representation of the synergistic relationship between drugs and cell lines in existing resistance prediction methods,this paper proposes a nonlinear subspace collaborative learning model,named NLS-DRP(Nonlinear Subspace-Driven Drug Resistance Prediction).The NLS-DRP consists of three key learning modules:the Cell branch,the Drug branch,and the Collaborative Fusion module.These modules are used to construct nonlinear subspaces for extracting cell line fea-tures,decompose drug structures to extract subsequence features,and design a nonlinear collaborative space for the fusion of cell line and drug features,respectively.Finally,by integrating the features from the three modules,the resistance of cell lines to drugs is predicted.Experiments conducted on two public datasets,the Genomics of Drug Sensitivity in Cancer(GDSC)and the Cancer Cell Line Encyclopedia(CCLE),demonstrate that the proposed NLS-DRP model significantly outperforms the benchmark methods,achieving a Pearson Correlation Coefficient(PCC)value of 0.945 8 and a Spearman's Correlation Coefficient(SCC)value of 0.924 2,thereby confirming the effectiveness of the method presented in our paper.