Use of “default” parameter settings when analyzing single cell RNA sequencing data using Seurat: a biologist’s perspective

Abstract

Aim: Analysis of large datasets has become integral to biological studies due to the advent of high throughput technologies such as next generation sequencing. Techniques for analyzing these large datasets are normally developed by bioinformaticists and statisticians, with input from biologists. Frequently, the end-user does not have the training or knowledge to make informed decisions on input parameter settings required to implement the analyses pipelines. Instead, the end-user relies on “default” settings present within the software packages, consultations with in-house bioinformaticists, or on methods described in previous publications. The aim of this study was to explore the effects of altering default parameters on the cell clustering solutions generated by a common pipeline implemented in the Seurat R package that is used to cluster cells based on single cell RNA sequencing (scRNAseq) data.

Methods: We systematically assessed the effect of altering input parameters by performing iterative analyses on a single scRNAseq dataset. We compared the clustering solutions using the different input parameters to determine which parameters have a large effect on cell clustering solutions.

Results: We used a range of input parameters for many, but not all, of the input parameters required by the Seurat R pipeline. We found that some input parameters had a very small effect on the clustering solution, while other parameters had a much larger effect.

Conclusion: We conclude that, when implementing the Seurat R package, the “default” parameters should be used with caution. We identified specific parameters that have a significant effect on clustering solutions.

Keywords

Single cell RNA sequencing / cell type annotation / Seurat R package / clustering algorithms

References

[1] RenF,ShiH,KanQ.Novel mechanisms and approaches to overcome multidrug resistance in the treatment of ovarian cancer..Biochim Biophys Acta2016;1866:266-75

[2] ZhengGX,BelgraderP.Massively parallel digital transcriptional profiling of single cells..Nat Commun2017;8:14049 PMCID:PMC5241818

[3] BrenneckeP,KimJK.Accounting for technical noise in single-cell RNA-seq experiments..Nat Methods2013;10:1093-5

[4] MereuE,MoutinhoC.Benchmarking single-cell RNA-sequencing protocols for cell atlas projects..Nat Biotechnol2020;38:747-55

[5] SahaiE,CukiermanE.A framework for advancing our understanding of cancer-associated fibroblasts..Nat Rev Cancer2020;20:174-86 PMCID:PMC7046529

[6] ThorssonV,BrownSD.The immune landscape of cancer..Immunity2019;51:411-2

[7] StuartT,HoffmanP.Comprehensive Integration of Single-Cell Data..Cell2019;177:1888-902.e21 PMCID:PMC6687398

[8] WooJ,StarrTK,WangJ.De novo prediction of cell-type complexity in single-cell RNA-seq and tumor microenvironments..Life Sci Alliance2019;2:e201900443 PMCID:PMC6607449

[9] RissoD,FletcherRB.clusterExperiment and RSEC: a bioconductor package and framework for clustering of single-cell and other large gene expression datasets..PLoS Comput Biol2018;14:e1006378 PMCID:PMC6138422

[10] KiselevVY,SchaubMT.SC3: consensus clustering of single-cell RNA-seq data..Nat Methods2017;14:483-6 PMCID:PMC5410170

[11] LinP,HoJW.CIDR: ultrafast and accurate clustering through imputation for single-cell RNA-seq data..Genome Biol2017;18:59 PMCID:PMC5371246

[12] AranD,LiuL.Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage..Nat Immunol2019;20:163-72 PMCID:PMC6340744

[13] HieB,NyquistSK,BergerB.Computational methods for single-cell RNA sequencing..Annu Rev Biomed Data Sci2020;3:339-64

[14] WaltmanL.A smart local moving algorithm for large-scale modularity-based community detection..Eur Phys J B2013;86:

[15] LytalN,AnL.Normalization methods on single-cell RNA-seq data: an empirical survey..Front Genet2020;11:41 PMCID:PMC7019105

[16] ScruccaL,MurphyTB.mclust 5: clustering, classification and density estimation using gaussian finite mixture models..R J2016;8:289-317 PMCID:PMC5096736