Background

Next-generation sequencing technology is enabling massive production of high-quality paired-end reads. Many platforms (Illumina Genome Analyzer, Applied Biosystems SOLID, Helicos HeliScope) are currently able to produce "ultra-short" paired reads of lengths starting at 25 nt. An analysis by Whiteford et al. [1] on sequencing using unpaired reads shows that ultra-short reads theoretically allow whole genome re-sequencing and de novo assembly of only small eukaryotic genomes. Chaisson, Brinza and Pevzner [2] recently determined that the paired read length threshold for de novo assembly of the E. coli genome is ≈ 35 nt, and ≈ 60 nt for the S. cerevisiae genome. The latter read length is unfeasible for some next-generation technologies. By conducting an analysis extending Whiteford et al. results, we investigate to what extent genome re-sequencing is feasible with ultra-short paired reads. We obtain theoretical read length lower bounds for re-sequencing that are also applicable to paired-end de novo assembly.

Methods

A novel algorithm that utilizes a suffix array has been specifically designed to compute the uniqueness of paired reads with fixed or variable mate-pair distance. The algorithm is a non-trivial extension of the RepAnalyse algorithm [3] to paired reads. Bacterial and eukaryotic genomes are analyzed to determine the uniqueness of paired reads given a fixed mate-pair distance of 300 nt. Longer mate-pair distances with high variability are also considered for the E. coli genome.

Discussion

Simulation results indicate that 97.4% of the E. coli genome is covered with unique paired reads of length 8 nt, and 90% of the H. sapiens genome is covered with unique paired reads of length 11 nt (see Figure 1). These results suggest that for large genomes, re-sequencing requires significantly shorter (for H. sapiens, at least 67% shorter) paired reads to achieve coverage comparable to unpaired reads. Moreover, a trade-off exists between read length and mate-pair distance: given a fixed mate-pair distance of 5,000 nt (resp. 2,000 nt), the whole E. coli genome can be unambiguously probed by paired reads of length above 18 nt (resp. 700 nt). When the uncertainty in mate-pair distance is ± 10%, only a small part of the genome cannot be uniquely probed (resp. 0.3% and 0.1% in the previous cases).

Figure 1
figure 1

Percentage of unique paired and unpaired reads as a function of read length for the E. coli and H. sapiens genomes. Paired uniqueness is computed with a mate-pair distance of 300 nt.

Full size image