Statistical Analysis Method Of Large Data Based On Linear Mixed Model And Its Application

Linear mixed models have become a kind of important statistical models. In recent years, linear mixed models have been widely used in various fields, and they can deal with multitudinous types of data.With the development of the information age, data-centric scientific research produced more and more large datasets, such as genomics, proteomics and brain science, etc. Linear mixed models were applied to high-throughput genomic or proteomics mass spectrometry dataset frequently.In order to solve the parameter estimation of linear mixed model, which design matrix is thin and tall, on regular desktop. Based on the idea of FaST-LMM, we extend the method used to get the dominant singular subspace. Combined the above method with the out of core notion we proposed two new methods to compute the Singular Value Decomposition of large thin and tall matrix and name the two methods as Column-oriented block R-SVD and Row-oriented block SVD. We combined the two methods with FaST-LMM to calculate the parameter estimation of linear mixed model of large data on ordinary computer. The simulations are given in the paper.In order to apply the FaST-LMM method to the analysis of melanoma dataset, one method is introduced to compute the SVD of block diagonal matrix in the paper. We analyze the data in R.Some valuable rules and conclusions are obtained via the research work of this paper.