The genome atlas: navigating a new era of reference genomes
Alice M. Kaye, Wyeth W. Wasserman
Abstract
Current approaches to generating a new reference genome are misguided because they often conflate the two distinct roles of the reference: (i) enabling knowledge exchange; and (ii) increasing computational efficiency. This dichotomy needs to be reflected in the underlying design of a new reference.There is widespread acceptance that the current linear reference genome needs to be expanded to incorporate the common variation observed across different populations for both equity and performance considerations.Switching to methods that index DNA reads rather than reference genomes offers advantages by preventing compatibility issues between reference versions. Indexed read sets have the longevity to adapt as sequencing technology advances, read length increases, and the volume of data being generated surges. The reference genome serves two distinct purposes within the field of genomics. First, it provides a persistent structure against which findings can be reported, allowing for universal knowledge exchange between users. Second, it reduces the computational costs and time required to process genomic data by creating a scaffold that can be relied upon by analysis software. Here, we posit that current efforts to extend the linear reference to a graph-based structure while trying to fulfil both of these purposes concurrently will face a trade-off between comprehensiveness and computational efficiency. In this article, we explore how the reference genome is used and suggest an alternative structure, The Genome Atlas (TGA), to fulfil the bipartite role of the reference genome. The reference genome serves two distinct purposes within the field of genomics. First, it provides a persistent structure against which findings can be reported, allowing for universal knowledge exchange between users. Second, it reduces the computational costs and time required to process genomic data by creating a scaffold that can be relied upon by analysis software. Here, we posit that current efforts to extend the linear reference to a graph-based structure while trying to fulfil both of these purposes concurrently will face a trade-off between comprehensiveness and computational efficiency. In this article, we explore how the reference genome is used and suggest an alternative structure, The Genome Atlas (TGA), to fulfil the bipartite role of the reference genome. The human reference genome has provided the foundation for years of genetic discovery and research, but recently, multiple review papers have highlighted the deficiencies of the linear reference, leading to a growing consensus that a richer reference structure is necessary for continued improvements in the era of widespread whole-genome sequencing (WGS) [1.Ballouz S. et al.Is it time to change the reference genome?.Genome Biol. 2019; 20: 159Crossref PubMed Scopus (51) Google Scholar, 2.Computational Pan-Genomics Consortium Computational pan-genomics: status, promises and challenges.Brief. Bioinform. 2018; 19: 118-135PubMed Google Scholar, 3.Yang X. et al.One reference genome is not enough.Genome Biol. 2019; 20: 104Crossref PubMed Scopus (22) Google Scholar]. The Human Genome Project published its first draft in 2001, which, as the only publicly available human genome at nucleotide resolution, became integrated as the backbone of genome analysis. The emergence of new techniques combined with the falling cost of WGS allowed each new release to resolve gaps, improving its overall stability and accuracy. The increased reliability of the reference genome, coupled with the desire for faster and less computationally intensive genome analysis to keep pace with rapidly evolving sequencing methods, shifted the focus of bioinformatics from genome assembly-based (see Glossary) to genome alignment-based algorithms. The availability of a persistent frame of reference enabled genetic variants and their functional roles to be cataloged using the base pair (bp) coordinate system in large databases, such as gnomAD [4.Karczewski K.J. et al.The mutational constraint spectrum quantified from variation in 141,456 humans.Nature. 2020; 581: 434-443Crossref PubMed Scopus (1713) Google Scholar] and ClinVar [5.Landrum M.J. et al.ClinVar: improving access to variant interpretations and supporting evidence.Nucleic Acids Res. 2018; 46: D1062-D1067Crossref PubMed Scopus (1028) Google Scholar]. Further improvements in computational methods, combined with long read sequencing, expanded such databases to include structural variations (SVs). Despite advances in computing, the size of WGS data forces the continued use of a reference scaffold against which to perform genome assembly, read alignment, and variant calling [6.Pereira R. et al.Bioinformatics and computational tools for next-generation sequencing analysis in clinical genetics.J. Clin. Med. Res. 2020; 9: 132Google Scholar,7.Lightbody G. et al.Review of applications of high-throughput sequencing in personalized medicine: barriers and facilitators of future progress in research and clinical application.Brief. Bioinform. 2019; 20: 1795-1811Crossref PubMed Scopus (39) Google Scholar]. Consequently, the reference genome has matured into an integral component of genome analysis pipelines. Over the past decade, the sharp decrease in sequencing costs has led to a flood of large-scale and population-specific sequencing projects, revealing unique sequences missing from the current reference genome [8.Sherman R.M. et al.Assembly of a pan-genome from deep sequencing of 910 humans of African descent.Nat. Genet. 2019; 51: 30-35Crossref PubMed Scopus (115) Google Scholar,9.Lee Y.-G. et al.Insertion variants missing in the human reference genome are widespread among human populations.BMC Biol. 2020; 18: 167Crossref PubMed Scopus (2) Google Scholar], population-specific differences in common genetic variants [10.Maretty L. et al.Sequencing and de novo assembly of 150 genomes from Denmark as a population reference.Nature. 2017; 548: 87-91Crossref PubMed Scopus (65) Google Scholar,11.Wong K.H.Y. et al.De novo human genome assemblies reveal spectrum of alternative haplotypes in diverse populations.Nat. Commun. 2018; 9: 3040Crossref PubMed Scopus (35) Google Scholar], and increasing recognition that some populations, particularly Indigenous, are at risk of being left behind [12.Garrison N.A. et al.Genomic research through an Indigenous lens: understanding the expectations.Annu. Rev. Genomics Hum. Genet. 2019; 20: 495-517Crossref PubMed Scopus (54) Google Scholar]. The GRCh38 reference assembly has been expanded to include alternate loci (ALT loci), most of which are similar to the primary assembly but contain many small variants that commonly occur together. Consequently, naively aligning to a concatenation of the reference genome and ALT loci results in reads that map to multiple ambiguous locations [13.Church D.M. et al.Extending reference assembly models.Genome Biol. 2015; 16: 13Crossref PubMed Scopus (88) Google Scholar]. Additionally, this approach is not an accurate representation of the underlying biological knowledge. It ignores the location information of ALT loci, their prevalence in different populations, and the link between reference sequences and ALT loci. This jury-rigged solution creates hurdles for traditional mapping algorithms [14.Church D.M. et al.Modernizing reference genome assemblies.PLoS Biol. 2011; 9e1001091Crossref PubMed Scopus (244) Google Scholar] and, while these can be partially overcome with newer methods [15.Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM.arXiv. 2013; 2013 (arXiv:1303.3997v2)Google Scholar], these methods often provide only minimal gains at the cost of space and time complexity. To better reflect the structure of the genome, focus has shifted to the concept of a pan-genome to capture the complete set of variations present in a an review of human R.M. in the human genome Rev. Genet. 2020; PubMed Scopus (54) Google Scholar]. and while the concept of a pan-genome is to its and be for the of genetic are to be for genome analysis. This that it is to use methods that reference genomes for et graph-based read against a approach and of current 2020; PubMed Scopus Google Scholar]. are a to such as de and have the of assembly the used for the first draft of the reference of the draft of the human genome with Res. PubMed Scopus Google Scholar]. The of computationally assembly algorithms for and an in the design of richer reference from the reference sequence and commonly observed have been used to the of read in et genome in the using a population reference Genet. 2015; PubMed Scopus Google Scholar] and algorithms et genome and with and 2019; PubMed Scopus Google Scholar, et of structural variation using Commun. 2019; PubMed Scopus Google Scholar, et pan-genome read to variation 2018; 19: PubMed Scopus Google Scholar, The across variant and using variant Genet. 2018; PubMed Scopus Google Scholar]. led by the for Genomics and has a of graph-based for genome analysis et with data a PubMed Scopus (22) Google Scholar]. many face common the for performance with linear The of a computationally reference genome that linear methods, while allowing using a coordinate to be In this article, we the to a from the of approaches and the role of the reference genome in the era of a of the a reference genome to the trade-off between computational and analysis the of The Genome Atlas (TGA), and which we will to the design of a new reference genome. current approaches to graph-based focus the linear reference into a are with sequences and haplotypes are as of the of graph-based reference by of the the as an for a representation of a pan-genome et and the of genome Res. 2017; PubMed Scopus Google Scholar]. of genome and Here, we many of the common genome how to and in a how to information such as and and how to while such as Pan-Genomics Consortium Computational pan-genomics: status, promises and challenges.Brief. Bioinform. 2018; 19: 118-135PubMed Google Scholar]. The et read mapping by genetic variation in the 2018; PubMed Scopus Google Scholar] highlighted by et as a of a genome has been to read mapping reads contain but it than the used linear This for a read better than methods et genome and with and 2019; PubMed Scopus Google G. et and accurate genomic using genome Genet. 2019; 51: PubMed Scopus Google Scholar], with et graph-based read against a approach and of current 2020; PubMed Scopus Google Scholar]. from the of variants a the of nucleotide variants and are to the increased of et variants for Biol. 2018; 19: PubMed Scopus Google Scholar]. This that using a pan-genome reference read mapping is not the current graph-based the index structure of a reference that is to computational efficiency. The primary of is to within a large set of data have been and for such approaches in et for large of sequencing 2019; by from in the R.M. in the human genome Rev. Genet. 2020; PubMed Scopus (54) Google et of structural variation using Commun. 2019; PubMed Scopus Google for multiple sequence S. PubMed Scopus Google Scholar]. these methods in size and time for large of genomes of Biol. Scholar], to be such as small variants H. et al.The design and of reference with Biol. 2020; PubMed Scopus Google Scholar]. a is as the of computational methods is to the primary focus of the representation for the a is approaches for will as has been the for the linear The in the current linear reference has a of an from the be as the approach of change is Genome reference that is only a to the current linear reference is to the of graph-based and will to the for in the To the longevity of a new reference, we to its role in the era of and a design to the of a with the to adapt as technology advances while for and upon The current reference genome has two but distinct to knowledge exchange and to the and of analysis algorithms To the between the two we can use the of a of knowledge of the genome, by the The the between which to be for a in a of and for between using such as can be within the of the but distinct a of that knowledge the human genome. In of the of a a a DNA sequence that is an a of this is The complete genome of an can be through two of read in a has two distinct sets of information with (i) and (ii) the The between and in genomic with such as and functional between and the within to the of we perform in the from for between within a for a we are to and for using of the by the that published within the which is the of with a that have a prevalence in a are using the the because this the The of for and the functional of the differences between genomes populations of genomes the of we have with to a reference to knowledge of the within a set of not we not to the of the to the in a The not two contain the and across in the be computationally the is the we the is for to the from the a of that knowledge the human genome. In of the of a a a DNA sequence that is an a of this is The complete genome of an can be through two of read in a has two distinct sets of information with (i) and (ii) the The between and in genomic with such as and functional This between and the within to the of we perform in the from for between within a for a we are to and for using of the by the that published within the which is the of with a that have a prevalence in a are using the the because this the The of for and the functional of the differences between genomes populations of genomes the of we have with to a reference genome. In to knowledge of the within a set of not we not to the of the to the in a The not two contain the and across in the be computationally the is the we the is for to the from the how the reference we that are two that are the of both assembly and (i) sequence a a and (ii) sequence from two different sequence is for of a sequence within a set of sequence between two sequences of for between within a reference genome and a between two reads The at which these can be and the of they be are the the of genome analysis algorithms. sequence upon data to The most data sets to use as an index are that change but that will be a large of an index of the genome is and sequences from read sets are for in the index to as read are diverse approaches to with some to index size et of DNA 2019; PubMed Scopus Google Scholar], time et of of and of index 2019; PubMed Scopus Google Scholar]. methods the reference genome being a structure It is to to an index structure it but for sequences it is often to the index we the of personalized and population-specific sequencing the reference genome is with personalized reference genomes accurate analysis results et graph-based read against a approach and of current 2020; PubMed Scopus Google Scholar]. The at which new genomic information is is posit to the most accurate analysis we to use the reference genome and databases available and, that these are being the reference is a data set to an index we the data sets to that are and for a set of the is an the between a reference genome and data from an The space and time for creating an reference genome is the most data sets to use as an index are that change but will be a large of In this the data set to be be as because it is the current reference genome is not the databases of genetic variations are growing rapidly to genome we the reference genome as knowledge of it can be a data set and, is an data set to use for an we a at the data sets in WGS we can that a read set generated for a genome is with the of and are and, it is to design an that is in time as the reads are by the sequencing S. a sequence 2018; Scopus Google advances in algorithms to the of a et S. of the and for and Scopus Google Scholar] have the to be for DNA sequences by that the of a a set of reads is unique the that the reads, provided that reads are through The of reads is not an because increasing read reduces the of reads to a and for the to be used as the between This to the of algorithms that each read with a small rather than as a which is often large to into the is an the between a reference genome and data from an The space and time for creating an reference genome is the most data sets to use as an index are that change but will be a large of In this the data set to be be as because it is the current reference genome is not the databases of genetic variations are growing rapidly to genome we the reference genome as knowledge of it can be a data set and, is an data set to use for an we a at the data sets in WGS we can that a read set generated for a genome is with the of and are and, it is to design an that is in time as the reads are by the sequencing S. a sequence 2018; Scopus Google Scholar]. advances in algorithms to the of a et S. of the and for and Scopus Google Scholar] have the to be for DNA sequences by that the of a a set of reads is unique the that the reads, provided that reads are through The of reads is not an because increasing read reduces the of reads to a and for the to be used as the between This to the of algorithms that each read with a small rather than as a which is often large to into that a read set is a data set for the reads are generated and to the read sequences not change genome analysis. the data reads of by a sequencing can be a data Additionally, because the reads are generated it be to design an that is in time S. a sequence 2018; Scopus Google Scholar]. a set of reads can be rapidly for sequences of such as structural of the traditional approach of from reads and in the reference genome to as for read sequences be from the reference and for within the read set to the The genome assembly and read algorithms is the of two This the of to the between two to for the that a read to a in the algorithms have been with improvements in from the of common variants in the reference rather than naively variants to the reference can to information and an in and et variants for Biol. 2018; 19: PubMed Scopus Google Scholar]. within this role of sequence it is that the of information within the reference genome be to only that information to the read set being The to and genomic findings across a information from diverse populations is to understanding of the human genome. with the role of the role of the reference genome in enabling knowledge exchange two (i) new and (ii) between of the genome. The of genome sequencing has led to databases of their functional and clinical most of these databases to use loci as the primary of a This to databases the as the underlying linear coordinate highlighted by the of the to a linear coordinate system between to in knowledge exchange are using the of the This to databases that the coordinate as the of a because to the and between versions. we to a structure that for to be and by than a coordinate while in clinical and research The knowledge exchange is the discovery of between The to and explore and in a is in the discovery of the functional of genetic data sets are with a of to required for of genomic of tools to genetic analysis and the of applications for genomic a to for the of new information with for data how the reference genome is we have the two roles exchange and computational and the primary of each and and To a trade-off between and computational it is to that a reference used for knowledge exchange not have to be the structure as the reference used as a computational for read and variant calling algorithms. This the to the that we are two reference for each The to be to the results of an analysis in a between two reference two reference is to be an solution because it the large of to a reference genome. we a representation of the reference genome that not a coordinate system to knowledge exchange can a genome, a graph-based representation of the personalized genome for a which provides a backbone for computational analysis and for the design of the new reference enabling knowledge exchange forces from the a coordinate system to a of the genome. each a unique of the genome, but not results from and has an data with minimal to between data can be of as in a which are through allowing to explore the genome through and and not by information can be with such as to between to such as can be as a allowing for information continued and into through can be through multiple a such as within a genome prevalence in a and This approach the knowledge and of and to be into the analysis process by the to in This is similar to and databases, such as but issues because it not use genomic as the primary for each the primary is a unique that we a used by the with to a used for This for information to be and for new and to in the The through that results and clinical findings new versions. be used to into the to the size of to with the of available and to be The use of for of to be with databases that contain such as information and This provide the to databases to the and data required for many WGS technology is read and current algorithms upon a reference genome to provide a to the computational of aligning large of this a reference genome can be of as a personalized genome, by a linear the most of variations present in an of an genome for an the of a of from of in as et the of sequence against reads with 2020; Google Scholar]. the genome from a of reduces its the time and space of algorithms. the read as for the to be and the of as the mapping process The of that link to the because variants present in the personalized genome are The focus of can be to the of variants through their to in The personalized genome can contain multiple which to the to a genome representation by two through the using between such as information to The provide a to the for the because only the needs to be the genome. This approach of that are common across multiple similar to only the for an space to be to the genetic to an index for a read set is the to the of the this is an of research, with new algorithms for index et and 2017; Scholar], use of et the the techniques to Biol. Bioinform. 2015; PubMed Scopus Google Scholar], and the to index reads in time as they are et variant calling of 2019; 9: PubMed Scopus (2) Google Scholar]. long read sequencing the current the advantages of will because reads will be et and analysis in small 2020; 16: Scopus Google Scholar] and the of will The in technology have to an the era of the linear is a of consensus as to which structure is for the era of reference with the longevity to through future The to a pan-genome that human information with is computationally for and and can be while is are many hurdles the of a new reference (see but we that a is required as to how we the role of a reference genome. The reference needs to be as an of information that both the and of knowledge we have the genome, a persistent for and has the for This from a reference to a of information forces the roles of the reference genome and read set to with the read set as the data set to be to the performance of This a genome from the information in to be used as the computationally backbone for information the human genome is being generated but between genome assembly is and can we to the reference genome with new information analysis while enabling the of with the current reference genome is the of reference the of sequences to the reference is often we the of human genetic variation in a reference to provide of genomic to current an index for the reference genome, a reference genome in the to an index for each sequencing we an to a index for a set of reads using this access to and performance by to perform genome increasing access to a of genetic variants for and but a reference genome with long of variants by can provide a to information in the of versions. are the to a persistent for each variant for human information the human genome is being generated but between genome assembly is and can we to the reference genome with new information analysis while enabling the of with the current reference genome is the of reference the of sequences to the reference is often we the of human genetic variation in a reference to provide of genomic to current an index for the reference genome, a reference genome in the to an index for each sequencing we an to a index for a set of reads using this access to and performance by to perform genome increasing access to a of genetic variants for and but a reference genome with long of variants by can provide a to information in the of versions. are the to a persistent for each variant for human of the and and for and the and for This by from the of Genome and and of alternative representation of a sequence for a genome that is to as a unique to of the genome in a similar to that between two include nucleotide variations in which a and in length are structural variations (SVs). of a genome by sequences to a reference genome to the from which the read of a genome by such as reads, into representation of a of sequences and are and, are in the a of length is a a is a of length such as a complete representation of the genetic variation present within a the of sequences and sequences that a genome.