Department of Computer Science and Engineering

1
Significance of Sequence Alignments

• Fall 2003
• Dr. Susan Bridges

2
Homologous sequences?

• When two proteins of nucleotides are very
  similar, biologists call them homologues.
• Homologues have the same ancestor, similar
  function, or similar structure.
• When we get an alignment score, we need to know
  the probability that the alignment between
  unrelated sequences would reach the score between
  two sequences of interest.

3
Orthologs, Paralogs and Homologs
Ancestral organism
X
Y
Speciation
Duplication
B
A
Y
Y
X
X
B
A
X1
X2
Ya
Yb
X1 and X2 are orthologs with same function.
Paralogs Ya and Yb may have different but related
functions.
Homologs
4
Homologues

• How similar must sequences be in order to be
  considered homologues?
• More than 25 percent of the amino acids present
  for proteins
• More than 70 of the nucleotides present for DNA
• Below the 25 and 70 limits is called the
  twilight zone
• These estimates only work for at least 100
  nucleotides or amino acids.

5
Distributions

• First efforts assumed that alignment scores would
  be normally distributed.
• This is not the case.
• Local alignment scores were found to follow the
  Gumbel Extreme Value Distribution
• BLAST alignments use a variation of this
  distribution called the Karlin-Altschul statistic

6
Distributions
7
Values Describing Scores

• Both the Gumbel Extreme Value Distribution and
  Karlin-Ashtul Distribution use E values and P
  values
• E-value (Expect value) the average number of
  times such a match would be found
• P-value (probability) probability of finding an
  alignment under assumptions
• Important note Alignments that are
  statistically important may not be biologically
  important

8
Evaluating Scores E Scores

• Goal is to find an alignment that is unlikely by
  chance
• The lower the E-value the better (unlikely to
  happen by chance).
• E-values close to 1 indicate that the alignment
  may not be significant because the probability it
  could happen by chance is high.

9
BLAST and E-values

• BLAST approximates E-values
• BLAST underestimates E-values (thinks they are
  better than they are).
• Rules of thumb for BLAST E-values
• E-value above 10-4 is not necessarily interesting
  (0.0001)
• To be certain of a homolog, E-value should be
  lower than 0.0001.