Impact of analytic provenance in genome analysis

Abstract

BackgroundMany computational methods are available for assembly and annotation of newly sequenced microbial genomes. However, when new genomes are reported in the literature, there is frequently very little critical analysis of choices made during the sequence assembly and gene annotation stages. These choices have a direct impact on the biologically relevant products of a genomic analysis - for instance identification of common and differentiating regions among genomes in a comparison, or identification of enriched gene functional categories in a specific strain. Here, we examine the outcomes of different assembly and analysis steps in typical workflows in a comparison among strains of Vibrio vulnificus.ResultsUsing six recently sequenced strains of V. vulnificus, we demonstrate the "alternate realities" of comparative genomics, and how they depend on the choice of a robust assembly method and accurate ab initio annotation. We apply several popular assemblers for paired-end Illumina data, and three well-regarded ab initio genefinders. We demonstrate significant differences in detected gene overlap among comparative genomics workflows that depend on these two steps. The divergence between workflows, even those using widely adopted methods, is obvious both at the single genome level and when a comparison is performed. In a typical example where multiple workflows are applied to the strain V. vulnificus CECT 4606, a workflow that uses the Velvet assembler and Glimmer gene finder identifies 3275 gene features, while a workflow that uses the Velvet assembler and the RAST annotation system identifies 5011 gene features. Only 3171 genes are identical between both workflows. When we examine 9 assembly/ annotation workflow scenarios as input to a three-way genome comparison, differentiating genes and even differentially represented functional categories change significantly from scenario to scenario.ConclusionsInconsistencies in genomic analysis can arise depending on the choices that are made during the assembly and annotation stages. These inconsistencies can have a significant impact on the interpretation of an individual genome's content. The impact is multiplied when comparison of content and function among multiple genomes is the goal. Tracking the analysis history of the data - its analytic provenance - is critical for reproducible analysis of genome data.

Kiwi genome provides insights into evolution of a nocturnal lifestyle

“…Unlike the majority of other birds analyzed so far, kiwi has a higher number of γ subgroup ORs. Gene family size estimates are highly dependent on genome quality [ 46 ] and continuous curation is ongoing even for well-annotated genomes: for example, in the chicken olfactory repertoire the number of annotated ORs changed by a factor of eight in two consecutive Ensembl releases (release 73 – 251 ORs and release 74 – 30 ORs). Further improvement of genome qualities, including kiwi, are therefore required for the identification of a complete set of ORs.…”

The NIST data science initiative

“…A genome sequence dataset may be recreated from raw data and the provenance records associated with genomic annotations [15].…”

Software Choice and Sequencing Coverage Can Impact Plastid Genome Assembly–A Case Study in the Narrow Endemic Calligonum bakuense

“…Several recent investigations have reported factors that may influence the accuracy of plastid genome assembly, including software choice (Freudenthal et al, 2020 ) and sequencing coverage (reviewed in Gruenstaeudl and Jenke, 2020 ). The choice of assembly software has been reported as a source of inconsistency in genome assembly by several previous studies (e.g., Magoc et al, 2013 ; Morrison et al, 2014 ). In the de novo assembly of plastid genomes from genome skimming data, such inconsistency may be associated with differences between assembly algorithms: while some software tools have implemented algorithms that conduct a cyclical sequence extension from a single “seed” sequence (e.g., Dierckxsens et al, 2017 ), others employ a kmer-based construction of contigs, followed by the concatenation of multiple contigs based on sequence overlap and similarity to a reference genome (e.g., Bakker et al, 2016 ; McKain and Wilson, 2017 ).…”