Research On Single-cell Dimensionality Reduction Algorithm Based On Manifold Learning

Single-cell RNA sequencing has great potential in discovering cell types,identifying cell states,tracking developmental lineages,and reconstructing cell spatial organization.With the development of single-cell sequencing technology,the results of sequencing have become more and more detailed,and more and more cells can be measured at a time.The single cell contains a wealth of information,in the process of analyzing and processing these data,effective dimensionality reduction becomes more important.This paper studies the currently popular single-cell data dimensionality reduction algorithm t-SNE,and applies the linear dimensionality reduction algorithm PCA and two classic dimensionality reduction algorithms,LLE and MDS,based on manifold learning to the single-cell data dimensionality reduction process.And through the analysis and research of the four algorithms,a combined dimensionality reduction method is proposed.First use PCA to reduce the dimension of the original data,and then use LLE,MDS and t-SNE to perform the second dimension reduction.This article first selects data from the GEO database,then filters and preprocesses the data,then uses the Linnorm normalization method to unify the data standards,and then performs dimensionality reduction clustering on the data set,and calculate the ARI value.Finally,by comparing the running time of the algorithm and the ARI value to judge the pros and cons of the algorithm.This article selects a public data set in the GEO database as the processing data set for empirical analysis.This data set contains six samples.This paper first uses four dimensionality reduction algorithms to reduce the dimensionality of these six samples,and compares the advantages and disadvantages of these four dimensionality reduction algorithms by the algorithm running time and ARI value.Then,for the problems of these algorithms,through the combination of algorithms,the effect of dimensionality reduction is improved,and the dimensionality reduction effects of the algorithms before and after combination are compared.Finally,the dimensionality reduction effects of the three combined algorithms are compared.From the results,it can be seen that the combined algorithm not only shortens the running time of the algorithm,but also improves the accuracy of the algorithm's dimensionality reduction,and has certain application value.