Genome similarity measures

[GerryB]

When a genome is fully sequenced, it can be compared to other genomes. For example, human and chimpanzees share 96% of their genome.

My question is what does this "96% similarity" really mean? I realized it's a boiled down summary for the popular press, but what's the actual measurement/comparison behind it, and what tools make these kinds of comparisons?

Does it mean that each genome has been run through a global alignment with each other and only 4% of the sequence has no reasonable match? (ha, is it even practical to consider a global alignment with a 4B bp genome? Are there any tools to even try? Would such a global alignment have any meaning since even the chromosome count differs?)

Or perhaps it means something like "if you take a random 100bp substring from the human genome, there's a 96% chance you can find the identical substring in the chimp genome." This makes some experimental and mathematical sense but in some ways it's poorly defined, since for example if you changed it to taking a random 5 bp substring from the genome, a human would have 100% dupication with an oak tree, since 5bp is so short it's easy to match.

In my particular example of chimp versus human, it was done by the Eichler Lab but it's unclear which of their papers talks about the comparison of genomes.

[Melissa]

This is an interesting story. Researchers also found out that there's a big difference between individuals in human (Forgot where I read this). So we might be closer to Chimp than each other!

This article put a misleading title just to attract attention but explains no further. Maybe they use the same approach as this group who claimed 98% similarity. Targeting only at certain chromosome. It was found out that chromosome 2 in human is the fusion of 2 small chromosomes in chimp. Found a nice article here explaining how it was measured when verifying the chromosome fusion issue. Perhaps you will be interested to look at comparative genomics to find out more.

I think it's complicated to compare two organism at the gene level with all the orthologous and paralogous genes problem. The common way to measure evolutionary relationship is by comparing the mitochondria or chloroplast sequences because they are well-conserved and don't follow Mendelian inheritance.

Hope this helps.

[cariaso]

http://en.wikipedia.org/wiki/Lies,_damned_lies,_and_statistics

There is no established right answer, and everyone seems to picked whatever metric makes their intended conclusion sound strongest.

[GerryB]

This article put a misleading title just to attract attention but explains no further. Maybe they use the same approach as this group who claimed 98% similarity. Targeting only at certain chromosome. It was found out that chromosome 2 in human is the fusion of 2 small chromosomes in chimp. Found a nice article here explaining how it was measured when verifying the chromosome fusion issue. Perhaps you will be interested to look at comparative genomics to find out more.

Wow, that's an excellent reference. (It's so funny that the article from the bible creationist website had quite nice detail!)
It does sound like for (part of one chromosome at least) they seem to use a global alignment and count indels and mismatches as a score of 0, matches as 1. Then you just take the global score and divide by the longer genome's length to reach a percentage.

And yes, the "proper" statistics for comparison are likely impossible to reduce down to a single number like 96%. Probably the subfield of comparative genomics has its own measurements and standards. The specific numerical chimp comparison by Britten looked at just 780K bp. I wonder if comparative geomics guys do any kind of genome-wide comparisons? What tools would even do such a thing with 1B bp?