Research On Dimension Reduction,Spatial Domain Detection And Trajectory Inference For Spatial Transcriptome Data

Traditional single-cell RNA sequencing(sc RNA-seq)technology can obtain cell sequence differences in specific microenvironments to facilitate the study of functional differences.However,because cells are enzymolized during the sequencing process,the spatial location information is not preserved.With the development of sequencing technology,the emergence of spatial transcriptome sequencing technology,which can obtain gene expression data while preserving location information,has facilitated the study of the spatial transcriptome landscape of cell tissues and new discoveries in many fields of biology.The data collected from various spatial transcriptome technologies not only have the same high-dimensional characteristics(gene number)as the general sc RNA-seq data,but also contain a large amount of spatial location information.Noise in the data also brings further difficulties to the analysis.Conventional dimensionality reduction methods applied to sc RNA-seq are much less effective in computing spatial transcriptome data because a large amount of spatial information is not utilized.In this paper,using spatial information,on the basis of dimensionality reduction and spatial domain identification,a new Spatial Walk model of pseudotime analysis and trajectory inferencetailored for spatial transcriptome was developed.The main work of this paper is as follows:Firstly,This paper combines the traditional dimensionality reduction method with Spatial PCA and the clustering method K-means and Walktrap algorithm to establish different spatial domain detecion models,using the spatial transcriptome dataset DLPFC human prefrontal cortex samples for analysis.And we select appropriate indicators to evaluate the performance of each model performance,so as to screen out the most suitable dimensionality reduction methods and clustering methods for spatial transcriptomics.The experimental results show that the spatial domain detection model built by Spatial PCA has a higher index score than the traditional dimensionality reduction method,indicating that the spatial location information has been fully utilized.Spatial PCA is an effective dimensionality reduction method for spatial transcription.Walktarp identifies spatial domains more accurately than K-means,based on which Walktrap was used for subsequent trajectory analysis.Secondly,this paper designs a spatial transcriptomics trajectory inference model Spatial Walk,which enables a new tailor-made downstream analysis for spatial transcriptomics.In this paper,the Walktrap algorithm was used to identify spatial domains and draw the minimum spanning tree through the clustering center of the spatial domain to generate the initial path of the trajectory,and the principal curve algorithm was used to iteratively fit the path into the data to get the smooth branching curve.Pseudotime was calculated by orthogonal projection,thus transforming the knowledge of the entire lineage structure into a robust estimation of the pseudotime variable.Another method of calculating pseudotime is given,which is based on diffusion map.Thirdly,DLPFC samples are used to test the performance of Spatial Walk.The experimental results show that Spatial Walk detects continuous smooth trajectories on tissues and pseudotime maps of high accuracy are constructed.The continuous pseudotime values between layers indicate the validity of Pseudotime calculated by SpatialWalk has good robustness.Spatial Walk reliably sorts cells along the inferred trajectory and successfully reveals the spatial internal structural connection of cortical cell organization.In conclusion,this study has designed a new analysis process for spatial transcriptomic data and evaluated traditional dimensionality reduction methods and spatial domain detection models.And on this basis,the Spatial Walk model for pseudotime analysis trajectory inference of spatial transcriptome was established,which opened up a way for future research.