Finishing the sequence of clones

Sequencing of large-insert clones began with production of an initial

assembly based on shotgun sequence data. For a typical BAC clone, we

generated 6�10-fold coverage in paired end-sequences from random,

small-insert (2�4 kb) plasmid clones and used computer programs (P.

Green, unpublished and see ref. 68) to assemble the data into sequence

contigs connected by linking information. Each base was assigned a

quality score reflecting its predicted accuracy, based on the underlying

shotgun sequence data. The assembly typically had gaps and lowquality

regions, with the number varying greatly across clones. These

regions are highly enriched in sequences that are difficult to clone or

sequence and thus are not represented even after deep (6�10-fold)

coverage with random reads. (These regions are also poorly covered by

whole-genome shotgun strategies69.)

The finishing phase converted this draft assembly into a high-quality

continuous sequence by obtaining directed information. It involved

iterative cycles of computational analysis and laboratory work. Box 2 Fig.

1 shows a simplified flowchart. The first step was to inspect the draft

assembly for evidence of mis-assembly, arising from inappropriate

merger of repeated sequences. Such evidence would include

inconsistent patterns of linking among contigs, regions with unusually

high coverage in sequence reads and bases with �high-quality

discrepancies� among the underlying sequence reads. In general,

sequence assembly is more straightforward for the clone-based

hierarchical shotgun strategy than for the whole-genome shotgun

strategy, because the use of clones avoids problems arising from

polymorphism and from different copies of repeated regions elsewhere

in the genome. Most clones passed assembly inspection, but some

failed due to the presence of very similar local dispersed, tandem or

inverted repeats. Careful inspection could resolve the problem in some

cases, but specific strategies had to be devised in other cases. One

approach was to isolate distinct copies of the repeat in subclones of

intermediate size (10-kb plasmids or fosmids) and sequence these

subclones. Box 2 Fig. 2 illustrates an initially mis-assembled BAC clone

from chromosome Y that could be assembled correctly with careful

editing.

The second step was gap closure. Because gaps tended to be

enriched for problematic sequences, gap closure was challenging; it

often required multiple attempts using a variety of alternative methods.

Gaps were classified into two types: �spanned� and �unspanned�.

Spanned gaps were those for which the two flanking contig ends were

linked by an end-sequenced plasmid. Most such gaps could be closed

by primer-directed sequencing of the plasmid, serially extending the