Alignment by numbers: sequence assembly using compressed numerical representations

MotivationDNA sequencing instruments are enabling genomic analyses of unprecedented scope and scale, widening the gap between our abilities to generate and interpret sequence data. Established methods for computational sequence analysis generally use nucleotide-level resolution of sequences, and while such approaches can be very accurate, increasingly ambitious and data-intensive analyses are rendering them impractical for applications such as genome and metagenome assembly. Comparable analytical challenges are encountered in other data-intensive fields involving sequential data, such as signal processing, in which dimensionality reduction methods are routinely used to reduce the computational burden of analyses. We therefore seek to address the question of whether it is possible to improve the efficiency of sequence alignment by applying dimensionality reduction methods to numerically represented nucleotide sequences.\n\nResultsTo explore the applicability of signal transformation and dimensionality reduction methods to sequence assembly, we implemented a short read aligner and evaluated its performance against simulated high diversity viral sequences alongside four existing aligners. Using our sequence transformation and feature selection approach, alignment time was reduced by up to 14-fold compared to uncompressed sequences and without reducing alignment accuracy. Despite using highly compressed sequence transformations, our implementation yielded alignments of similar overall accuracy to existing aligners, outperforming all other tools tested at high levels of sequence variation. Our approach was also applied to the de novo assembly of a simulated diverse viral population. Our results demonstrate that full sequence resolution is not a prerequisite of accurate sequence alignment and that analytical performance can be retained and even enhanced through appropriate dimensionality reduction of sequences.