Ariadne: Synthetic Long Read Deconvolution Using Assembly Graphs

Abstract

Background: De novo assemblies are critical for capturing the genetic composition of complex samples. Linked-read sequencing techniques such as 10x Genomics' Linked-Reads, UST's TELL-Seq, Loop Genomics' LoopSeq, and BGI's Long Fragment Read combines 30 barcoding with standard short-read sequencing to expand the range of linkage resolution from hundreds to tens of thousands of base-pairs. The application of linked-read sequencing to genome assembly has demonstrated that barcoding-based technologies balance the ffs between long-range linkage, per-base coverage, and costs. Linkedreads come with their own challenges, chief among them the association of multiple long fragments with the same 30 barcode. The lack of a unique correspondence between a long fragment and a barcode, in conjunction with low sequencing depth, confounds the assignment of linkage between short-reads. Results: We introduce Ariadne, a novel linked-read deconvolution algorithm based on assembly graphs, that can be used to extract single-species read-sets from a large linked-read dataset. Ariadne deconvolution of linked-read clouds increases the proportion of read clouds containing only reads from a single fragment by up to 37.5-fold. Using these enhanced read clouds in de novo assembly significantly improves assembly contiguity and the size of the largest aligned blocks in comparison to the non-deconvolved read clouds. Integrating barcode deconvolution tools, such as Ariadne, into the postprocessing pipeline for linked-read technologies increases the quality of de novo assembly for complex populations, such as microbiomes. Ariadne is intuitive, computationally efficient, and scalable to other large-scale linked-read problems, such as human genome phasing.

QuickDeconvolution: fast and scalable deconvolution of linked-read sequencing data

“…On the ATCC loopseq dataset, Minerva proposed a deconvolution for <0.05% of the reads, as already reported in Mak et al (2021) and was thus not evaluated. Ariadne and QD classified more than 99% of the reads.…”

“…Using the terminology defined in previous papers ( Danko et al , 2019 ; Mak et al , 2021 ), the set of reads sharing the same barcode will be referred to as a read cloud . The barcodes provide implicit long-range information: two reads sharing the same barcode originate with high probability from the same fragment and are thus ‘not far away’ on the DNA strand.…”

“…Very recently, another reference-free software from the authors of Minerva has been proposed under the name Ariadne ( Mak et al , 2021 ). Based on a totally different concept, Ariadne starts by doing a complete assembly of the de-Bruijn graph of reads using the SPAdes assembler ( Prjibelski et al , 2020 ), ignoring the barcodes.…”