ProSolo: Accurate Variant Calling from Single Cell DNA Sequencing Data

Abstract

1Obtaining accurate mutational profiles from single cell DNA is essential for the analysis of genomic cell-to-cell heterogeneity at the finest level of resolution. However, sequencing libraries suitable for genotyping require whole genome amplification, which introduces allelic bias and copy errors. As a result, single cell DNA sequencing data violates the assumptions of variant callers developed for bulk sequencing, which when applied to single cells generate significant numbers of false positives and false negatives. Only dedicated models accounting for amplification bias and errors will be able to provide more accurate calls.We present ProSolo, a probabilistic model for calling single nucleotide variants from multiple displacement amplified single cell DNA sequencing data. It introduces a mechanistically motivated empirical model of amplification bias that improves the quantification of genotyping uncertainty. To account for amplification errors, it jointly models the single cell sample with a bulk sequencing sample from the same cell population-also enabling a biologically relevant imputation of missing genotypes for the single cell.Through these innovations, ProSolo achieves substantially higher performance in calling and genotyping single nucleotide variants in single cells in comparison to all state-of-the-art tools. Moreover, ProSolo implements the first approach to control the false discovery rate reliably and flexibly; not only for single nucleotide variant calls, but also for artefacts of single cell methodology that one may wish to identify, such as allele dropout. ProSolo's model is implemented into a flexible framework, encouraging extensions. The source code and usage instructions are available at: https://github.com/prosolo/prosolo ADO -allele dropout, alt -alternative, err -error, het -heterozygous, hom -homozygous, ref -reference. 6/28 18/28

Scelestial: fast and accurate single-cell lineage tree inference based on a Steiner tree approximation algorithm

“…For optimal performance, many of these probabilistic approaches require substantial time for optimization (e.g., using MCMC). The increasing sample sizes of single-cell datasets emphasizes the need for fast and scalable methods in this domain that also maintain high accuracy [29].…”

Accurate genotyping of single cells with Octopus

“…We ran Octopus (v0.7.4), SCcaller (v2.0.0), MonoVar (v0.0.1), and Prosolo (v0.6.1) in a similar way to a previous analysis of this dataset 15 . For Octopus, we used the cell calling model to jointly call all cells and bulk clones C1-3.…”

“…Similar to previous work [11][12][13][14][15][16] , we designed a calling model for Octopus that accounts for amplification stochasticity in scWGA data (Methods). However, unlike other single-cell methods, Octopus discovers candidate variants with local de novo assembly; realigns reads to candidate haplotypes to account for sequencing and alignment errors; leverages physical read linkage information during genotyping; and recalibrates genotype quality scores using machine learning.…”

“…However, none of the calling models previously discussed are appropriate for genotyping single cells as they do not account for common sources of uncertainty found in single-cell whole-genome amplified (scWGA) sequencing data, such as amplification stochasticity that can result in allelic dropout (ADO) 9 . Indeed, due to these artifacts, accurate variant calling of scWGA data is still considered a "grand challenge" of single-cell technology 10 .Similar to previous work [11][12][13][14][15][16] , we designed a calling model for Octopus that accounts for amplification stochasticity in scWGA data (Methods). However, unlike other single-cell methods, Octopus discovers candidate variants with local de novo assembly; realigns reads to candidate haplotypes to account for sequencing and alignment errors; leverages physical read linkage information during genotyping; and recalibrates genotype quality scores using machine learning.…”

“…To assess calling accuracy, we ran Octopus on previously studied human primary fibroblast cells with closely related bulk samples that provide approximate ground truth for the cellular genotypes 12,14,15 (Methods). Octopus had considerably higher genotyping accuracy than SCcaller 12 -the only other single cell method that calls indels -despite SCcaller being given the baseline heterozygous variants to fit its allele bias model (Fig.…”