Giters

shenyang1981 / scRNA-seq_notes

A list of scRNA-seq analysis tools

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

scRNA-seq data analysis tools and papers

MIT License PR's Welcome

Single-cell RNA-seq related tools and genomics data analysis resources. Tools are sorted by publication date, newest on top. Unpublished tools are listed at the end of each section. Please, contribute and get in touch! See MDmisc notes for other programming and genomics-related notes.

Table of content

Preprocessing pipelines

  • All steps in scRNA-seq analysis, QC (count depth, number of genes, % mitochondrial), normalization (global, downsampling, nonlinear), data correction (batch, denoising, imputation), feature selection, dimensionality reduction (PCA, diffusion maps, tSNE, UMAP), visualization, clustering (k-means, graph/community detection), annotation, trajectory inference (PAGA, Monocle), differential analysis (DESeq2, EdgeR, MAST), gene regulatory networks. Description of the bigger picture at each step, latest tools, their brief description, references. R-based Scater as the full pipeline for QC and preprocessing, Seurat for downstream analysis, scanpy Python pipeline. Links and refs for tutorials. https://github.com/theislab/single-cell-tutorial

  • kallistobus - fast pipeline for scRNA-seq processing. New BUS (Barcode, UMI, Set) format for storing and manipulating pseudoalignment results. Includes RNA velocity analysis. Python-based

  • PyMINEr - Python-based scRNA-seq processing pipeline. Cell type identification, detection of cell type-enriched genes, pathway analysis, co-expression networks and graph theory approaches to interpreting gene expression. Notes on methods: modified K++ clustering, automatic detection of the number of cell types, co-expression and PPI networks. Input: .txt or .hdf5 files. Detailed analysis of several pancreatic datasets

  • dropEst - pipeline for pre-processing, mapping, QCing, filtering, and quantifying droplet-based scRNA-seq datasets. Input - FASTQ or BAM, output - an R-readable molecular count matrix. Written in C++

  • Scanpy - Python-based pipeline for preprocessing, visualization, clustering, pseudotime and trajectory inference, differential expression and network simulation

  • SEQC - Single-Cell Sequencing Quality Control and Processing Software, a general purpose method to build a count matrix from single cell sequencing reads, able to process data from inDrop, drop-seq, 10X, and Mars-Seq2 technologies

  • bigSCale - scalable analytical framework to analyze large scRNA-seq datasets, UMIs or counts. Pre-clustering, convolution into iCells, final clustering, differential expression, biomarkers.Correlation metric for scRNA-seq data based on converting expression to Z-scores of differential expression. Robust to dropouts. Matlab implementation. Data, 1847 human neuronal progenitor cells

  • zUMIs - scRNA-seq processing pipeline that handles barcodes and summarizes  UMIs using exonic or exonic + intronic mapped reads (improves clustering, DE detection). Adaptive downsampling of oversequenced libraries. STAR aligner, Rsubread::featureCounts counting UMIs in exons and introns.

  • demuxlet - Introduces the ‘demuxlet’ algorithm, which enables genetic demultiplexing, doublet detection, and super-loading for droplet-based scRNA-seq. Recommended approach when samples have distinct genotypes

  • CALISTA - clustering, lineage reconstruction, transition gene identification, and cell pseudotime single cell transcriptional analysis. Analyses can be all or separate. Uses a likelihood-based approach based on probabilistic models of stochastic gene transcriptional bursts and random technical dropout events, so all analyses are compatible with each other. Input - a matrix of normalized, batch-removed log(RPKM) or log(TPM) or scaled UMIs. Methods detail statistical methodology. Matlab and R version

  • scPipe - A preprocessing pipeline for single cell RNA-seq data that starts from the fastq files and produces a gene count matrix with associated quality control information. It can process fastq data generated by CEL-seq, MARS-seq, Drop-seq, Chromium 10x and SMART-seq protocols. Modular, can swap tools like use different aligners

  • RAPIDS & Scanpy Single-Cell RNA-seq Workflow - real-time analysis of scRNA-seq data on GPU. Tweet

  • sceasy - An R package to convert different single-cell data formats to each other, supports Seurat, SingleCellExperiment, AnnData, Loom

  • MAESTRO - Model-based AnalysEs of Single-cell Transcriptome and RegulOme - a comprehensive single-cell RNA-seq and ATAC-seq analysis suit built using snakemake

  • STAR alignment parameters: –outFilterType BySJout, –outFilterMultimapNmax 100, –limitOutSJcollapsed 2000000 –alignSJDBoverhangMin 8, –outFilterMismatchNoverLmax 0.04, –alignIntronMin 20, –alignIntronMax 1000000, –readFilesIn fastqrecords, –outSAMprimaryFlag AllBestScore, –outSAMtype BAM Unsorted. From Azizi et al., “Single-Cell Map of Diverse Immune Phenotypes in the Breast Tumor Microenvironment.”

Visualization pipelines

scATAC-seq

  • Benchmarking of 10 scATAC-seq analysis methods (brief description of each in Methods) on 10 synthetic (various depth and noise levels) and 3 real datasets. scATAC technology overview, problems. Three clustering methods (K-means, Louvain, hierarchical clustering), adjusted Rand index, adjusted mutual information, homogeneity for benchmarking against gold-standard clustering, Residual Average Gini Index for benchmarking against gene markers (silver standard). SnapATAC, Cusanovich2018, cisTopic perform best overall. R code, Jupyter notebooks https://github.com/pinellolab/scATAC-benchmarking/

  • scATAC-pro - pipeline for scATAC-seq mapping, QC, peak detection, clustering, TF and GO enrichment analysis, visualization (via VisCello). Compared with Scasat, Cellranger-atac.

  • ArchR - R package for processing and analyzing single-cell ATAC-seq data. Compared to Signac and SnapATAC, has more functionality, faster. Input - BAM files. Efficient h5-based storage allows for large dataset processing. Doublet detection, genome-wide 500bp binning and peak identification, assignment to genes using best performing model, dimensionality reduction, clustering. Code to reproduce the paper, https://github.com/GreenleafLab/ArchR

  • scOpen - estimating open chromatin status in scATAC-seq experiments, aka imputation/smoothing of extreme sparse matrices. Uses positive-unlabelled learning of matrices to estimate the probability that a region is open in a given cell. The probability matrix can be used as input for downstream analyses (clustering, visualization). Integrated with the footprint transcription factor activity score (scHINT). scOpen estimated matrices tested as input for scABC, chromVAR, cisTopic, Cicero, improve performance

  • SnapATAC - scATAC-seq pipeline for processing, clustering, and motif identification. Genome is binned into equal-size (5kb) windows, binarized with 1/0 for ATAC reads present/absent, Jaccard similarity between cells, normalized to account for sequencing depth (observed over expected method, two others), PCA on the matrix KNN graph and Louvain clustering to detect communities, tSNE or UMAP for visualization. Motif analysis and GREAT functional enrichment for each cluster. Outperforms ChromVAR, LSA, Cicero, Cis-Topic. Very fast, can be applied to ChIP-seq, scHi-C. https://github.com/r3fang/SnapATAC

  • scABC, single-cell Accessibility Based Clustering - scATAC-seq clustering. Weights cells by a nonlinear transformation of library sizes, then, weighted K-medoids clustering. Input - single-cell mapped reads, and the full set of called peaks. Applied to experimental and synthetic scATAC-seq data, outperforms simple K-means-based clustering, SC3

  • Cicero - connect distal regulatory elements with target genes (covariance-based, graphical Lasso to compute a regularized covariance matrix) along pseudotime-ordered (Monocle2 or 3) scATAC-seq data. Optionally, adjusts for batch covariates. Applied to the analysis of skeletal myoblast differentiation, sciATAC-seq. R package, https://cole-trapnell-lab.github.io/cicero-release/

  • ChromVAR - scATAC-seq analysis. Identifying peaks, get a matrix of counts across aggregated peaks, tSNE for clustering, identifying motifs. Integrated with Seurat. https://github.com/GreenleafLab/chromVAR

Quality control

Normalization

Batch effect, merging

Imputation

Assessment of 18 scRNA-seq imputation methods (model-based, smooth-based, deep learning, matrix decomposition). Similarity of scRNA- and bulk RNA-seq profiles (Spearman), differential expression (MAST and Wilcoxon), clustering (k-means, Louvain), trajectory reconstruction (Monocle 2, TSCAN), didn't test velocity. scran for normalization. Imputation methods improve correlation with bulk RNA-seq, but have minimal effect on downstream analyses. MAGIC, kNN-smoothing, SAVER perform well overall. Plate- and droplet-derived scRNA-seq cell line data, Additional File 4), Summary table of the functionality of all imputation methods, Additional File 5 - Hou, Wenpin, Zhicheng Ji, Hongkai Ji, and Stephanie C. Hicks. “A Systematic Evaluation of Single-Cell RNA-Sequencing Imputation Methods.” Genome Biology 21, no. 1 (December 2020)

Dimensionality reduction

Clustering

Spatial inference

Time, trajectory inference

Networks

RNA velocity

Differential expression

CNV

Annotation, subpopulation identification

Cell markers

Simulation

  • scDesign - scRNA-seq data simulator and statistical framework to access experimental design for differential gene expression analysis. Gamma-Normal mixture model better fits scRNA-seq data, accounts for dropout events (Methods describe step-wise statistical derivations). Single- or double-batch sequencing scenarios. Comparable or superior performance to simulation methods splat, powsimR, scDD, Lun et al. method. DE tested using t-test. Applications include DE methods evaluation, dimensionality reduction testing. https://github.com/Vivianstats/scDesign

  • Splatter - scRNA-seq simulator and pre-defined differential expression. 6 methods, description of each. Issues with scRNA-seq data - dropouts, zero inflation, proportion of zeros, batch effect. Negative binomial for simulation. No simulation is perfect. https://github.com/Oshlack/splatter

Power

Benchmarking

Deep learning

Spatial transcriptomics

scATAC-seq integration, Multi-omics methods

  • Single-cell ATAC + RNA co-assay methods - overview of technologies and protocols, references to the original papers

  • Multi-omics methods - Table 1 from Sierant, Michael C., and Jungmin Choi. “Single-Cell Sequencing in Cancer: Recent Applications to Immunogenomics and Multi-Omics Tools.” Genomics & Informatics 16, no. 4 (December 2018)

  • Signac is an extension of Seurat for the analysis, interpretation, and exploration of single-cell chromatin datasets, and integration with scRNA-seq. ChromatinAssay object class, Latent Semantic Indexing and the modified TF-IDF procedure for dimensionality reduction. Applied to the PBMC 10X multiomics dataset and the Brain Initiative Cell Census Network data. Also, the Sinto Python package for processing aligned single-cell data

  • UnionCom - integration of multi-omics single-cell data using unsupervised topological alignment. Based on GUMA (generalized unsupervised manifold alignment) algorithm. Three steps: 1) embedding each single-cell dataset into the geometric distance matrix; 2) Align distance matrices; 3) Project unmatched features onto common embedding space. Tested on simulated and experimental data (sc-GEM, scNMT). Neighborhood overlap metric for testing, outperforms Seurat, MMD-MA, scAlign.

  • scAI - integrative analysis of scRNA-seq and scATAC-seq or scMethylation data measured from the same cells (in contrast to different measures sampled from the same cell population). Overview of multi-omics single-cell technologies, methods for data integration in bulk samples and single-cell samples (MATCHER, Seural, LIGER), sparsity (scATAC-seq is ~99% sparse and nearly binary). Deconvolution of both single-cell matrices into gene loading and locus loading matrices, a cell loading matrix, in which factors K correspond to loadings of gene, locus, and cell in the K-dimensional space. A strategy to reduce over-aggregation. Cell subpopulations identified by Leiden clustering of the cell loading matrix. Visualization of the low-rank matrices with the Sammon mapping. Multi-omics simulation using MOSim, eight scenarios of simulated data, AUROC and Normalized Mutual Information (NMI) assessment of matrix reconstruction quality. Compared with MOFA, Seurat, LIGER. Tested on 8837 mammalian kidney cells scRNA-seq and scATAC-seq data, 77 mouse ESCs scRNA-seq and scMethylation, interpretation. https://github.com/sqjin/scAI

  • Harmony - scRNA-seq integration by projecting datasets into a shared embedding where cells differences between cell clusters are maximized while differences between datasets of origin are minimized = focus on clusters driven by cell differences. Can account for technical and biological covariates. Can integrate scRNA-seq datasets obtained with different technologies, or scRNA- and scATAC-seq, scRNA-seq with spatially-resolved transcriptomics. Local inverse Simpson Index (LISI) to test for database- and cell-type-specifc clustering. Outperforms MNN, BBKNN, MultiCCA, Scanorama. Memory-efficient, fast, scales to large datasets, included in Seurat. https://github.com/immunogenomics/harmony, Python version

  • Seurat v.3 paper. Integration of multiple scRNA-seq and other single-cell omics (spatial transcriptomics, scATAC-seq, immunophenotyping), including batch correction. Anchors as reference to harmonize multiple datasets. Canonical Correlation Analysis (CCA) coupled with Munual Nearest Neighborhoors (MNN) to identify shared subpopulations across datasets. CCA to reduce dimensionality, search for MNN in the low-dimensional representation. Shared Nearest Neighbor (SNN) graphs to assess similarity between two cells. Outperforms scmap. Extensive validation on multiple datasets (Human Cell Atlas, STARmap mouse visual cortex spatial transcriptomics. Tabula Muris, 10X Genomics datasets, others in STAR methods). Data normalization, variable feature selection within- and between datasets, anchor identification using CCA (methods), their scoring, batch correction, label transfer, imputation. Methods correspond to details of each Seurat function. Preprocessing of real single-cell data. https://satijalab.org/seurat/, https://github.com/satijalab/Integration2019

    • Stuart, Tim, Andrew Butler, Paul Hoffman, Christoph Hafemeister, Efthymia Papalexi, William M Mauck, Marlon Stoeckius, Peter Smibert, and Rahul Satija. “Comprehensive Integration of Single Cell Data.” Preprint. Genomics, November 2, 2018.

  • MAESTRO - MAESTRO (Model-based AnalysEs of Single-cell Transcriptome and RegulOme) is a comprehensive single-cell RNA-seq and ATAC-seq analysis suit built using snakemake. https://github.com/liulab-dfci/MAESTRO, Tweet

10X Genomics

10X QC

Data

Human

Cancer

  • CancerSEA - cancer scRNA-seq studies. Download individual studies, as well as gene signatures (from Angiogenesis, DNA damage to EMT, metastasis, etc.)

  • Azizi, Elham, Ambrose J. Carr, George Plitas, Andrew E. Cornish, Catherine Konopacki, Sandhya Prabhakaran, Juozas Nainys, et al. “Single-Cell Map of Diverse Immune Phenotypes in the Breast Tumor Microenvironment.” Cell, June 2018. - scRNA-seq of immune cells in BRCA - continuous activation of T cells, no macrophage polarization. inDrop and 10X platforms. 47,016 CD45+ cells from 8 primary breast carcinomas. 83 clusters, tested by cross-validation. Data 1, Data 2, Data 3

Mouse

Brain

Links

Papers

About

A list of scRNA-seq analysis tools

License:MIT License

Languages

Language:R 100.0%