Biochem 503, Fall 2003

Protein sequence comparison and Protein evolution

• My questions concern the ability of software to recognize low complexity regions, and remove them from the sequence comparison.
  1. Does it matter which amino acid(s) are repeated in the region of low complexity? E.G. Tryptophan is a relatively rare amino acid, and Leucine is a relatively abundant amino acid in proteins. Ergo, if there is a region of amino acids that is 33% tryptophan vs. 33% leucine, would the program recognize that tryptophan is rare and that its presence may be significant in determining homologous proteins in contrast to the leucine?

     Yes, amino-acid frequencies are taken into account. For more information, see Wootton and Federhen (1993) Comput. Chem. 17:149-163.

  2. Will the software count related amino acids as a region of low complexity? If there is a area high of V, L, I; D and E; or S and T can the software recognize that the region has say 50% S/T content (but only 25% serine) and notably exclude that region as a low complexity region?

     I'm not certain, but I think not. The assumption is that the process that produces low complexity regions is not necessarily the "normal" mutation process; seg/pseg focusses more on identities.

  3. Can you use the Smith-Waterman algorithm and use different matrices (PAM 250/PAM 40)

     Yes - one can use any of the PAM or BLOSUM matrices

• As you said, homologous proteins need not share even detectable sequence similarity and vs. there maybe exist high-scoring alignment while being unrelated and having completely different structure. Then I think we can draw no conclusion from the similarity comparition because high score alignment is not reliable and low score( no detectable sequence similarity?) alignment maybe is a homolog.

  You are mistaken. There are no high-scoring alignments that are unrelated with different structures. Sequences that share significant similarity (because they have high alignment scores) always have very similar structures, as structures change more slowly than sequences. High-scoring alignments are very reliable.

• I know that an E value is just the probability of aligning a given sequence at random against a query sequency times the number of the number of given sequences in a db. Here is were I'am a little confused. How exactly is the probability which ranges from 0 The formula for the probability P is given in the handout from the extreme value distribution: p(S') = 1 - exp(-exp(-S')) where S' is the normalized score. The scoring matrix comes into play in calculating the score, and also in the "lambda" parameter used to normalize the score (S' = lambda S - ln K m n).

Biochem 503 Home page