Single-cell Transcriptome Reconstruction With Compressed Sensing Strategy

With the development of high-throughput sequencing technology,people's understanding of relationship between gene expression and phenotypes has been increased.While traditional transcriptome sequencing can only give an average level of gene expression in a specific tissue or cell population and masks the differences in single cell level,single-cell transcriptome sequencing provides more comprehensive information.Commonly used single-cell RNA sequencing(sc RNA-seq)technologies can be divided into two main categories according to their library-construction methods,which are plate-based methods and microdroplet-based methods.Both of them have their own advantages,but they also have some limitations.The former methods are labor-intensive,costly,and time-consuming;the latter methods have relatively lower sensitivity and lose source information of single cells.How to balance the relationship between library construction cost and sequencing sensitivity,to acquire transcriptome data with high sensitivity and low cost when preserving source information of each cell,is a challenge for developing sc RNA-seq,which has great theoretical significance and application value.This thesis supplements and further improves the single-cell transcriptome compressed sequencing method on the basis of our preliminary work.Based on the compressed sensing theory,our method utilizes the sparse nature of single-cell transcriptome to compress and sequence single cell data simultaneously.According to designed measurement matrix,this method subsamples and mixes single-cell transcriptome to obtain mixed pools,and then sequences mixed pools to reconstruct the original single-cell gene expression profile.When being able to preserve source information of each cell,this method can also save costs and experiment time due to the decrease of sequencing libraries.Both computer simulations and actual sequencing experiment revealed that this method has relatively high sensitivity,and the reconstructed transcriptome has high consistency with the sequencing results of conventional sc RNA-seq method.The main contents of this thesis are as follows:1.Construction and Optimization of Single Cell Transcriptome Compressed Sequencing MethodIn this thesis,we elucidate the specific scheme of applying compressed sensing theory to single cell transcriptome compressed sequencing method,including experiment protocols and data analysis process.The supplement and improvement of our preliminary work are mainly reflected in:(1)When processing the mixed pool sequencing data in the actual situation,normalization of sequencing depth should be conducted according to the number of single cell samples added in each mixed pool;(2)We apply different regularization models to reconstruct samples of different sparsity thus improve reconstruction effect.2.Reconstruction of Single-Cell Transcriptome on Computer SimulationsWe used two single-cell transcriptome datasets with different sparsity to verify the reconstruction effect of Basis Pursuit model and Ridge Regression model.Dimension reduction based on t-SNE and k-means clustering are used to analyze the reconstructed data.The simulation results revealed that this method can preserve relatively high sensitivity of gene detection when subsampling times are sufficient.Reconstruction effect based on Basis Pursuit model are better when applied to data of higher sparsity,but it is relatively easy to be affected by subsampling times and turbulence level;Reconstruction based on Ridge Regression model has better effect on the data of lower sparsity,and it has relatively high calculation speed and robustness to turbulence.3.Experimental Verification of Single-Cell Transcriptome Compressed Sequencing MethodIn this thesis,seven kinds of human immune cell lines were cultured as experimental samples.The single cells' transcriptomes were reverse transcribed,amplified and purified by kit based on Smart-Seq2 method to obtain cDNA.cDNA samples were then subsampled according to computer-generated compressed sensing measurement matrix to acquire 40 mixed pools.We used the sequencing data of mixed pools and measurement matrix to reconstruct single-cell transcriptome and compared it with result acquired from traditional sc RNA-seq method.The experiment demonstrated that the gene detection sensitivity of our method can reach 86.46% of traditional Smart-Seq2 method's sensitivity.Due to the high similarity and low data sparsity of our samples,the reconstruction effect of Ridge Regression model is significantly better than that of Basis Persuit model(P<0.0001).The reconstructed gene expression data are in high consistency with the result of traditional sc RNA-seq,and the Pearson correlation coefficient is 0.891.Based on turbulence model of our method,we evaluated the turbulence in this experiment.The estimated turbulence is 0.31?0.35,which is of great significance to predict the disturbance level of transcriptome sequencing data in the future.