Comparative Analysis of Single-Cell RNA Sequencing Methods

Slides:



Advertisements
Similar presentations
Pol II Docking and Pausing at Growth and Stress Genes in C. elegans
Advertisements

Xiaoshu Chen, Jianzhi Zhang  Cell Systems 
Figure 1. Effect of acute TNF treatment on transcription in human SGBS adipocytes as assessed by RNA-seq and RNAPII ChIP-seq. Following 10 days in vitro.
Assessing Copy Number Alterations in Targeted, Amplicon-Based Next-Generation Sequencing Data  Catherine Grasso, Timothy Butler, Katherine Rhodes, Michael.
Alternative Computational Analysis Shows No Evidence for Nucleosome Enrichment at Repetitive Sequences in Mammalian Spermatozoa  Hélène Royo, Michael Beda.
Adaptive Evolution of Gene Expression in Drosophila
Global Mapping of Human RNA-RNA Interactions
CEL-Seq: Single-Cell RNA-Seq by Multiplexed Linear Amplification
Christopher R. Cabanski, Vincent Magrini, Malachi Griffith, Obi L
Volume 65, Issue 2, Pages (January 2017)
Volume 54, Issue 1, Pages (April 2014)
Transient N-6-Methyladenosine Transcriptome Sequencing Reveals a Regulatory Role of m6A in Splicing Efficiency  Annita Louloupi, Evgenia Ntini, Thomas.
Shiran Bar, Maya Schachter, Talia Eldar-Geva, Nissim Benvenisty 
Reliable Identification of Genomic Variants from RNA-Seq Data
SAGA Is a General Cofactor for RNA Polymerase II Transcription
Pejman Mohammadi, Niko Beerenwinkel, Yaakov Benenson  Cell Systems 
Hyeshik Chang, Jaechul Lim, Minju Ha, V. Narry Kim  Molecular Cell 
A Massively Parallel Reporter Assay of 3′ UTR Sequences Identifies In Vivo Rules for mRNA Degradation  Michal Rabani, Lindsey Pieper, Guo-Liang Chew,
Volume 25, Issue 6, Pages e4 (November 2018)
Adrien Le Thomas, Georgi K. Marinov, Alexei A. Aravin  Cell Reports 
The Translational Landscape of the Mammalian Cell Cycle
Volume 6, Issue 5, Pages (May 2010)
Volume 66, Issue 1, Pages e5 (April 2017)
Target-Specific Precision of CRISPR-Mediated Genome Editing
Volume 68, Issue 5, Pages e7 (December 2017)
Volume 3, Issue 1, Pages (July 2016)
Volume 85, Issue 4, Pages (February 2015)
Integrative Multi-omic Analysis of Human Platelet eQTLs Reveals Alternative Start Site in Mitofusin 2  Lukas M. Simon, Edward S. Chen, Leonard C. Edelstein,
Joseph Rodriguez, Jerome S. Menet, Michael Rosbash  Molecular Cell 
Biased Allelic Expression in Human Primary Fibroblast Single Cells
Hyeshik Chang, Jaechul Lim, Minju Ha, V. Narry Kim  Molecular Cell 
Volume 5, Issue 4, Pages e4 (October 2017)
Modeling Enzyme Processivity Reveals that RNA-Seq Libraries Are Biased in Characteristic and Correctable Ways  Nathan Archer, Mark D. Walsh, Vahid Shahrezaei,
Volume 23, Issue 1, Pages 9-22 (January 2013)
Volume 14, Issue 7, Pages (February 2016)
Volume 63, Issue 4, Pages (August 2016)
Taming Human Genetic Variability: Transcriptomic Meta-Analysis Guides the Experimental Design and Interpretation of iPSC-Based Disease Modeling  Pierre-Luc.
Volume 4, Issue 5, Pages e5 (May 2017)
Volume 54, Issue 5, Pages (June 2014)
Revisiting Global Gene Expression Analysis
Volume 88, Issue 4, Pages (November 2015)
Michal Levin, Tamar Hashimshony, Florian Wagner, Itai Yanai 
Volume 6, Issue 2, Pages e5 (February 2018)
A Clinically Validated Diagnostic Second-Generation Sequencing Assay for Detection of Hereditary BRCA1 and BRCA2 Mutations  Ian E. Bosdet, T. Roderick.
Pol II Docking and Pausing at Growth and Stress Genes in C. elegans
Jong-Eun Park, Hyerim Yi, Yoosik Kim, Hyeshik Chang, V. Narry Kim 
Volume 64, Issue 6, Pages (December 2016)
Volume 26, Issue 5, Pages (March 2016)
Shiran Bar, Maya Schachter, Talia Eldar-Geva, Nissim Benvenisty 
Baekgyu Kim, Kyowon Jeong, V. Narry Kim  Molecular Cell 
Volume 7, Issue 3, Pages e12 (September 2018)
Volume 47, Issue 2, Pages (July 2012)
Volume 6, Issue 2, Pages e5 (February 2018)
Alterations in mRNA 3′ UTR Isoform Abundance Accompany Gene Expression Changes in Human Huntington’s Disease Brains  Lindsay Romo, Ami Ashar-Patel, Edith.
Volume 50, Issue 2, Pages (April 2013)
Volume 53, Issue 6, Pages (March 2014)
Volume 42, Issue 6, Pages (June 2011)
Distal Alternative Last Exons Localize mRNAs to Neural Projections
Yuichiro Mishima, Yukihide Tomari  Molecular Cell 
Volume 9, Issue 3, Pages (November 2014)
Volume 10, Issue 2, Pages (August 2011)
Modeling Enzyme Processivity Reveals that RNA-Seq Libraries Are Biased in Characteristic and Correctable Ways  Nathan Archer, Mark D. Walsh, Vahid Shahrezaei,
Differential protein, mRNA, lncRNA and miRNA regulation by p53.
Brandon Ho, Anastasia Baryshnikova, Grant W. Brown  Cell Systems 
Julia K. Nussbacher, Gene W. Yeo  Molecular Cell 
Manfred Schmid, Agnieszka Tudek, Torben Heick Jensen  Cell Reports 
Xiaoshu Chen, Jianzhi Zhang  Cell Systems 
Volume 11, Issue 7, Pages (May 2015)
Development of a Novel Next-Generation Sequencing Assay for Carrier Screening in Old Order Amish and Mennonite Populations of Pennsylvania  Erin L. Crowgey,
Volume 27, Issue 7, Pages e4 (May 2019)
Presentation transcript:

Comparative Analysis of Single-Cell RNA Sequencing Methods Christoph Ziegenhain, Beate Vieth, Swati Parekh, Björn Reinius, Amy Guillaumet-Adkins, Martha Smets, Heinrich Leonhardt, Holger Heyn, Ines Hellmann, Wolfgang Enard  Molecular Cell  Volume 65, Issue 4, Pages 631-643.e4 (February 2017) DOI: 10.1016/j.molcel.2017.01.023 Copyright © 2017 Elsevier Inc. Terms and Conditions

Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions

Figure 1 Schematic Overview of the Experimental and Computational Workflow Mouse embryonic stem cells (mESCs) cultured in 2i/LIF and ERCC spike-in RNAs were used to generate single-cell RNA-seq data with six different library preparation methods (CEL-seq2/C1, Drop-seq, MARS-seq, SCRB-seq, Smart-seq/C1, and Smart-seq2). The methods differ in the usage of unique molecular identifier (UMI) sequences, which allow the discrimination between reads derived from original mRNA molecules and duplicates generated during cDNA amplification. Data processing was identical across methods, and the given cell numbers per method and replicate were used to compare sensitivity, accuracy, precision, power, and cost efficiency. The six scRNA-seq methods are denoted by color throughout the figures of this study as follows: purple, CEL-seq2/C1; orange, Drop-seq; brown, MARS-seq; green, SCRB-seq; blue, Smart-seq; and yellow, Smart-seq2. See also Figures S1 and S2. Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions

Figure 2 Schematic Overview of Library Preparation Steps For details, see the text. See also Table S1. Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions

Figure 3 Sensitivity of scRNA-Seq Methods (A) Percentage of reads (downsampled to one million per cell) that cannot be mapped to the mouse genome (gray) are mapped to regions outside exons (orange) or inside exons (blue). For UMI methods, dark blue denotes the exonic reads with unique UMIs. (B) Median number of genes detected per cell (counts ≥1) when downsampling total read counts to the indicated depths. Dashed lines above one million reads represent extrapolated asymptotic fits. (C) Number of genes detected (counts ≥1) per cell. Each dot represents a cell and each box represents the median and first and third quartiles per replicate and method. (D) Cumulative number of genes detected as more cells are added. The order of cells considered was drawn randomly 100 times to display mean ± SD (shaded area). See also Figures S3 and S4. Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions

Figure 4 Accuracy of scRNA-Seq Methods ERCC expression values (counts per million reads for Smart-seq/C1 and Smart-seq2 and UMIs per million reads for all others) were correlated to their annotated molarity. Shown are the distributions of correlation coefficients (adjusted R2 of linear regression model) across methods. Each dot represents a cell/bead and each box represents the median and first and third quartiles. See also Figure S5. Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions

Figure 5 Precision of scRNA-Seq Methods We compared precision among methods using the 13,361 genes detected in at least 25% of all cells by any method in a subsample of 65 cells per method. (A) Distributions of dropout rates across the 13,361 genes are shown as violin plots, and medians are shown as bars and numbers. (B) Extra Poisson variability across the 13,361 genes was calculated by subtracting the expected amount of variation due to Poisson sampling (square root of mean divided by mean) from the CV (SD divided by mean). Distributions are shown as violin plots and medians are shown as bars and numbers. For 349, 336, 474, 165, 201, and 146 genes for CEL-seq2/C1, Drop-seq, MARS-seq, SCRB-seq, Smart-seq/C1, and Smart-seq2, respectively, no extra Poisson variability could be calculated. See also Figures S6 and S7. Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions

Figure 6 Power of scRNA-Seq Methods Using the empirical mean/dispersion and mean/dropout relationships (Figures S8A and S8B), we simulated data for two groups of n cells each for which 5% of the 13,361 genes were differentially expressed, with log-fold changes drawn from observed differences between microglial subpopulations from a previously published dataset (Zeisel et al., 2015). The simulated data were then tested for differential expression using limma (Ritchie et al., 2015), from which the average true positive rate (TPR) and the average false discovery rate (FDR) were calculated (Figure S9A). (A) TPR for one million reads per cell for sample sizes n = 16, n = 32, n = 64, n = 128, n = 256, and n = 512 per group. Boxplots represent the median and first and third quartiles of 100 simulations. (B) TPR for one million reads per cell for n = 64 per group with and without using UMI information. Boxplots represent the median and first and third quartiles of 100 simulations. (C) TPRs as in (A) using mean/dispersion and mean/dropout estimates from one million (as in A), 0.5 million, and 0.25 million reads. Line areas indicate the median power with SE from 100 simulations. See also Figures S8–S10 and Table 1. Molecular Cell 2017 65, 631-643.e4DOI: (10.1016/j.molcel.2017.01.023) Copyright © 2017 Elsevier Inc. Terms and Conditions