Theoretical Bounds on Mate-Pair Information for Accurate Genome Assembly

TL;DR

This paper establishes theoretical bounds on the number of mate-pair libraries needed for accurate genome assembly, showing it is feasible with a logarithmic number under certain conditions, and provides insights into the problem's complexity.

Contribution

It offers the first theoretical analysis quantifying the number of mate-pair libraries required for accurate genome assembly based on repetitive region length.

Findings

Accurate assembly requires roughly R/2L mate-pair libraries in worst-case scenarios.

A simple polynomial-time algorithm can assemble the genome with (R/L)+1 libraries.

Under certain conditions, only O(log(R/L)) libraries are needed for guaranteed correctness.

Abstract

Over the past two decades, a series of works have aimed at studying the problem of genome assembly: the process of reconstructing a genome from sequence reads. An early formulation of the genome assembly problem showed that genome reconstruction is NP-hard when framed as finding the shortest sequence that contains all observed reads. Although this original formulation is very simplistic and does not allow for mate-pair information, subsequent formulations have also proven to be NP-hard, and/or may not be guaranteed to return a correct assembly. In this paper, we provide an alternate perspective on the genome assembly problem by showing genome assembly is easy when provided with sufficient mate-pair information. Moreover, we quantify the number of mate-pair libraries necessary and sufficient for accurate genome assembly, in terms of the length of the longest repetitive region within a genome. In our analysis, we consider an idealized sequencing model where each mate-pair library generates pairs of error free reads with a fixed and known insert size at each position in the genome. Even in this idealized model, we show that accurate genome reconstruction cannot be guaranteed in the worst case unless at least roughly R/2L mate-pair libraries are produced, where R is the length of the longest repetitive region in the genome and L is the length of each read. On the other hand, if (R/L)+1 mate-pair libraries are provided, then a simple algorithm can be used to find a correct genome assembly easily in polynomial time. Although (R/L)+1 mate-pair libraries can be too much to produce in practice, the previous bounds only hold in the worst case. In our last result, we show that if additional conditions hold on a genome, a correct assembly can be guaranteed with only O(log (R/L)) mate-pair libraries.