Novel Definition and Algorithm for Chaining Fragments with Proportional Overlaps

Abstract

Abstract. Chaining fragments is a crucial step in genome alignment. Existing chaining algorithms compute a maximum weighted chain with no overlaps allowed between adjacent fragments. In practice, using local alignments as fragments, instead of MEMs, generates frequent overlaps between fragments, due to combinatorial reasons and biological factors, i.e. variable tandem repeat structures that differ in number of copies between genomic sequences. In this paper, in order to raise this limitation, we formulate a novel definition of a chain, allowing overlaps proportional to the fragments lengths, and exhibit an efficient algorithm for computing such a maximum weighted chain. We tested our algorithm on a dataset composed of 694 genome couples and accounted for significant improvements in terms of coverage, while keeping the running times below reasonable limits.

YOC, A new strategy for pairwise alignment of collinear genomes

“…Although it is not the ultimate genome aligner, we argue that in practice, YOC combines important advantages: Simplicity of the algorithm: only similarity detection and chaining are performed, which avoids including badly aligned regions. Simplicity of use: as the spaced seeds are already optimized for bacterial genomes, YOC only requires the tuning of two parameters: (i) the E-value threshold for YASS, the higher the better if the goal is to ensure high sensitivity regardless of the level of divergence, and (ii) the overlap ratio for the chaining algorithm even though, as shown in [ 35 ], OC results are highly robust with respect to this parameter. MGA, MAUVE, and ProgressiveMAUVE include additional parameters linked to the four phases strategy, for instance the lengths of the matches that are used in the first and the third phases (P1 and P3) are critical.…”

“…In [ 35 ], we argued that the difficulty of using local alignments is that the chances that two adjacent fragments overlap are much higher than with short matches. At that point, we observed that such overlaps are commonly due to randomness, to methodological reasons during the fragment computation phase, or to biological phenomena, like tandem repeats.…”

“…Thus, accepting overlaps regardless of the fragment lengths is not the right solution. To get round this limitation, we extended the classical framework of the MWC in [ 35 ], by authorizing overlaps between fragments in the computed chain. We formalized the Maximum Weighted Chain with Proportional Length Overlap problem, where overlaps are proportional to the length of adjacent fragments.…”

“…The algorithm is based on a box representation of a trapezoid graph [ 38 ], with an adaptation of the sweep line paradigm to this problem. In [ 35 ], the OC tool was tested on real data and compared to classical chainers with respect to simple quantitative measures, and its robustness was proved with respect to its only parameter, the overlap ratio (default value = 10%). In YOC, the tool presented here, we rely on OverlapChainer (OC) for the chaining phase.…”

“…The YOC strategy can be described as follows. Phase (1): YOC enhances the similarity detection phase by computing local alignments with YASS [ 26 ]; and phase (2): it chains the local alignments using a recent chaining algorithm, OverlapChainer (OC) [ 35 ]. As it relies on YASS, a pairwise local aligner, and OC, a collinear chaining method, YOC is designed for the alignment of collinear regions of genome pairs.…”