Gene Feature Extraction Based On Multi-regularized Constraints And Low-rank Matrix Factor

Gene expression profile is an important research content of modern medicine,it can help to know about patient’s condition quickly by analyzing and processing gene expression profile data,which will provide an effective reference for subsequent treatment.Different from conventional data,gene expression profile data has two characteristics.Firstly,the original data generally has high dimension.There are a large number of redundant features and heavy noise.Secondly,the number of samples available for research is very small.These characteristics have brought great challenges for gene expression profile’s analysis and processing.The rank of the matrix is a mathematical representation of the data information’s complexity.Low-rank matrix factor is a commonly used model in the field of pattern recognition.And it is widely used in subspace recovery and classification as its good data recovery capabilities.Aiming at the technical problems in the process of gene expression profile’s analysis,this paper has done the following works which references the traditional low-rank matrix factor model and the classical theory in manifold learning.1.Analysing the current situation and related theories of gene expression profile.The related concepts,processing steps and related feature extraction methods of gene expression profile are introduced.The related theories of low-rank matrix factor and manifold learning are described in detail.The principles of some classifiers used in the paper are presented.2.Aiming at high dimension and heavy noise of gene expression profile,a dimension reduction model based on multi-regularized constraints and nonnegative matrix factor is proposed.Referring to the advantage of the nonnegative matrix factor model in data dimension reduction,the low-rank sparse constraint and manifold regularized constraint are introduced on this basis.The nonnegative matrix factor is used to acquire the low dimensional feature of high dimensional gene expression profile.The nearest neighbor structure of the original data is preserved by the manifold regularized constraints.At the same time,the low-rank sparse constraint is used to remove the noise of the gene expression profile data.The feasibility of the algorithm is proved by the classification performance of the extracted features under the classifier.3.In order to recovery the subspace of gene expression profile under the condition of smallsample,a feature extraction method of gene expression profile based on nonnegative dual graph regularized latent low-rank representation is proposed.The subspace data recovery ability of the gene expression profile under small-sample is enhanced by latent low-rank representation which observes the main feature and the latent feature of the data at the same time.The dual graph regularized which maintains the neighbor relationship structure of the original data are introduced on the basis of traditional latent low-rank representation model.The noise is removed effectively by referencing the low-rank sparse constraints in computer vision.Nonnegative constraint gives the calculation a certain sparsity,and gives the algorithm better robustness.Compared with the ordinary low-rank representation and latent low-rank representation model,the proposed nonnegative dual graph regularized latent low-rank representation model has better feature extraction performance for small-sample gene expression profile data.