基于稀疏正则低秩张量回归的基因组数据分析

GENOME DATA ANALYSIS BASED ON SPARSE REGULARIZED LOW RANK TENSOR REGRESSION

为确保基因组数据分析中系数张量稀疏化,从而保证稳定精确的估计结果,提出一种基于稀疏正则低秩张量回归分析算法。利用张量协变量的结构,对张量数据进行特征选择;基于回归系数张量的Tucker分解,将超高维数据进行降维处理;进一步对因子矩阵施加正交约束,并设计一种基于交替方向乘法器算法的高效优化算法求解模型。对合成数据以及黑色素瘤基因组数据的分析表明,该算法具有较好的预测性能,并能识别出具有重要意义的标志物。

In order to ensure the sparseness of coefficient tensors and the stable and accurate estimation results,a method of genome data analysis based on sparse regularized low rank tensor regression is proposed.The structure of tensor co-variate was used to select features of tensor data.Based on Tucker decomposition of regression coefficient tensor,the dimension of super high dimensional data was reduced.The sparse penalty was directly applied to the coefficient tensor and an efficient optimization algorithm model based on alternating direction multiplier algorithm was designed.The analysis of synthetic data and melanoma genome data shows that the model has good prediction performance and can identify important markers.