FORGe: prioritizing variants for graph genomes

Abstract

There is growing interest in using genetic variants to augment the reference genome into a "graph genome" to improve read alignment accuracy and reduce allelic bias. While adding a variant has the positive effect of removing an undesirable alignmentscore penalty, it also increases both the ambiguity of the reference genome and the cost of storing and querying the genome index. We introduce methods and a software tool called FORGe for modeling these effects and prioritizing variants accordingly. We show that FORGe enables a range of advantageous and measurable trade-offs between accuracy and computational overhead. more variation to the reference eventually reduces alignment accuracy. We suggest efficient models for scoring variants according to the effect on accuracy and "blowup" (computational overhead), and further show that these scores can be used to achieve a balance of accuracy and overhead superior to current approaches. For example, extrapolating to a whole-human DNA sequencing experiment at 40-fold average coverage, we estimate that a well-engineered augmented reference can yield about 4.8M more correctly aligned reads and 1.2M fewer incorrectly aligned compared to the linear reference. Our methods for selecting variants also reduce reference bias, a chief goals of graph genomes. Finally, we compare the accuracy yielded by our methods to that achieved using an ideal personalized graph genome. We show that our methods approach the ideal much more closely than both linear genomes -even when they are modified to contain only major allelesand graph genomes built on different sets of variants.These methods are implemented in a new open source software tool called FORGe. We demonstrate FORGe in conjunction with the HISAT2 [12] graph aligner and with another aligner based on the Enhanced Reference Genome [7]. But FORGe's models and methods are suitable for any aligner that can include variants in the reference.

Recovery of non-reference sequences missing from the human reference genome

“…There is growing interest in using genetic variants to augment the reference genome into a graph genome [18][19][20]. To create a representative graph genome, the full spectrum of structural variations, including the alternate alleles, should be understood clearly.…”

Recovery of non-reference sequences missing from the human reference genome

“…There is growing interest in using genetic variants to augment the reference genome into a graph genome (Crysnanto et al, 2019;Pritt et al, 2018;Rakocevic et al, 2019). To create a representative graph genome, the full spectrum of structural variations, including the alternate alleles, should be understood clearly.…”

Reducing reference bias using multiple population reference genomes

“…We also assessed the alignment methods using an ideal, diploid personalized reference genome (Section 3.2). Results using the personalized reference serve as a rough upper bound on what is achievable with references that lack foreknowledge of donor genotypes 17,25,26 . We call this a "rough" upper bound because, while the personalized reference is ideal in that it contains the correct variants, the accuracy of alignment is also affected by tool-specific heuristics.…”

“…While graph aligners [11][12][13][14][15] can reduce reference bias, linear aligners still perform better on certain classes of reads 16 and graph-aligner performance is sensitive to the number of variants considered 17 . Other efforts have focused on elaborating the linear alignment paradigm to address reference bias.…”

“…Some studies suggest replacing the typical linear reference with a "major-allele" version, with each variant set to its most common allele. This can increase alignment [16][17][18] and genotyping accuracy 19 . The major-allele reference is largely compatible with the standard reference (though indels can shift coordinates) and imposes little or no additional computational overhead.…”