On the sequencing of the human genome

Abstract

Two recent papers using different approaches reported draft sequences of the human genome. The international Human Genome Project (HGP) used the hierarchical shotgun approach, whereas Celera Genomics adopted the whole-genome shotgun (WGS) approach. Here, we analyze whether the latter paper provides a meaningful test of the WGS approach on a mammalian genome. In the Celera paper, the authors did not analyze their own WGS data. Instead, they decomposed the HGP's assembled sequence into a "perfect tiling path", combined it with their WGS data, and assembled the merged data set. To study the implications of this approach, we perform computational analysis and find that a perfect tiling path with 2-fold coverage is sufficient to recover virtually the entirety of a genome assembly. We also examine the manner in which the assembly was anchored to the human genome and conclude that the process primarily depended on the HGP's sequence-tagged site maps, BAC maps, and clone-based sequences. Our analysis indicates that the Celera paper provides neither a meaningful test of the WGS approach nor an independent sequence of the human genome. Our analysis does not imply that a WGS approach could not be successfully applied to assemble a draft sequence of a large mammalian genome, but merely that the Celera paper does not provide such evidence.

Figure 1

Sequencing strategies. (Left) The hierarchical shotgun (HS) strategy involves decomposing the genome into a tiling path of overlapping BAC clones, performing shotgun sequencing on and reassembling each BAC, and then merging the sequences of adjacent clones. The method has the advantage that all sequence contigs and scaffolds derived from a BAC belong to a single compartment with respect to anchoring to the genome. (Right) Whole-genome shotgun (WGS) strategy involves performing shotgun sequencing on the entire genome and attempting to reassemble the entire collection. With the WGS method, each contig and scaffold is an independent component that must be anchored to the genome. In general, many scaffolds may not be anchored without directed efforts. (Contigs are contiguous blocks of sequence; scaffolds are sets of contigs joined by paired reads from both ends of a plasmid insert.)

Figure 2

Random vs. perfect spacing in 100 kb. (Upper) Two-fold coverage in perfectly spaced reads. (Lower) Two-fold coverage in randomly selected reads. There are frequent regions in which the adjacent reads either fail to overlap or the overlap is too small to allow reliable detection (< 40 bp). These breaks in continuity are indicated by vertical lines.