[Bioperl-l] E-value of a combined alignment?

[Ian Korf]

There are several publications on combined statistical significance of 
local alignment scores. The ones implemented in BLAST are not exactly 
the same as the publications though. You can get a pretty decent 
approximation by subtracting log(KMN) for each gap, but this isn't the 
proper formula.

WU-BLAST is much better for combined statistics than NCBI-BLAST because 
it shows the actual groups with the -links parameter and allows you to 
limit the number of groups with the -topcomboN and -topcomboE 
parameters. It also lets you fine-tune the groupings a bit with -olmax 
and -olfmax. If the sequences aren't too diverged, you might be better 
off keeping X low though.

-Ian

On Wednesday, September 3, 2003, at 12:06 AM, Yee Man Chan wrote:

>
> Hi folks,
>
> 	I am aligning mRNAs against human genome using ungapped tblastx. I
> got a bunch of HSPs with different e-values. I can observed that some 
> of
> them should be in the same group because they are exons of a gene. But
> then what is the e-value of all these HSPs combined?
>
> 	I know the formulas of e-value and bit score for BLOSUM62:
>
> Let S' be bit score, S be score, e be e-value, m be the length of HSP,
> n be length of database.
>
> S' = (0.318 * S - ln(0.135)) / ln(2)
>
> e = m * n / (2^(S'))
>
> 	I am guessing the formula for the e-value of
> non-overlapping combined e-value to be:
>
> S'' = (0.318 * sum_of_S - ln(0.135)) / ln(2)
>
> e' = sum_of_m * n / (2^(S''))
>
> 	Is this correct? Or do you know the right way to calculate it?
>
> Thanks in advance.
> Yee Man
>
> _______________________________________________
> Bioperl-l mailing list
> Bioperl-l at portal.open-bio.org
> http://portal.open-bio.org/mailman/listinfo/bioperl-l
>