Coverage requirement for denovo sequencing

[seqqeq]

Is there any study to explore the relationship between read length, raw read accuracy, coverage, genome size and the resulting number of contigs for denovo assembly? I am particularly interested in human genome. For example, if we have 10k reads, do we need less throughput to achieve the same quality assembly?

Appreciate someone give me some pointers. Simulation is fine too.

[krobison]

A place to start would be Lander-Waterman statistics

http://www.math.ucsd.edu/~gptesler/186/slides/shotgun_12-handout.pdf

however, these don't incorporate the actual structure of the human genome -- notably the repeats. These will favor long reads even more.

[seqqeq]

A place to start would be Lander-Waterman statistics

http://www.math.ucsd.edu/~gptesler/186/slides/shotgun_12-handout.pdf

however, these don't incorporate the actual structure of the human genome -- notably the repeats. These will favor long reads even more.

Thanks for the pointer. Its a very useful set of notes. In the NIH estimate for human genome sequencing cost:

The following 'sequence coverage' values were used in calculating the cost per genome:

Sanger-based sequencing (average read length=500-600 bases): 6-fold coverage
454 sequencing (average read length=300-400 bases): 10-fold coverage
Illumina and SOLiD sequencing (average read length=50-100 bases): 30-fold coverage

(http://www.genome.gov/sequencingcosts/)

They listed different coverage numbers from that estimated from Lander-Waterman statistics. If I use genome size 3Gb, 1 contig, read lengths 600, 400 and 100 bps, the formula gives coverage of 20.2, 18.7 and 18.3. (The numbers above are 30, 10 and 6). Any idea why its way off, esp for Sanger?

[jebowers]

I would assume the NIH estimates take into account repeats and error rates as well as sequencing biases.

Different sequencing technologies favor different GC% sequences, and to get the 99% coverage some technologies must sequence more to get the needed coverage of GC% extremes.

Also repeats complicate matters, as short sequences containing repeats can not be effectively anchored, requiring higher coverage.

Finally error rates result in some loss of coverage, again depending on the tech used.

So in summary, yes a formula would be great, but guidelines based on experimental results are probably going to be more accurate as several different factors besides read length strongly impact the results.