Notes 20100921 McConkey Bioinformatics BIOL 614

Jump to: navigation, search

Content

discussing BLAST
see the two papers posted today
statistics are important in this field!
- see: http://www.statsoft.com/textbook/
statistics concepts reviewed today ...
- normal distribution (Gaussian)
- μ (average)
- σ (standard deviation)
- n; (sample size)
- p-value; (probability that a datapoint belongs to a different distribution)

we get p-value by taking the area under the gaussian and only count the area after a given data point
are two samples from the same population: student's t-test
BLAST: each returned sequence in the result can be...
- an actual match for the query sequence
- a non-match for the query sequence (a false positive)
the distribution isn't actually Gaussian; instead, there is a long right-handed tail (is skew left).
alignments and gap penalties: gap penalties are discovered empirically

search space -- BLOSUM and Smith-Waterman are simply too slow for this
BLAST and FASTA was thus developed; it's a lot faster
heuristics employed -- approximation, not guaranteed to find best result; only retrieves a fraction of the search space
taxonomy can be returned instead of just p-values, e-values
graphical descriptions -- notice that the query is returned with all of the parameters echoed
looking for a potassium channel protein
bit scores
E-values ...
- E > 1 -- chance alignment likely
- E < 0.05 or 0.1 -- represents biological significance

word building (these are 3mers)
find word matches between two sequences
seed expansion in both directions
1990: each word match was extended
1997: two hits needed before seeds would be expanded (must be a certain distance between one another; can have gaps between them)
BLAST follows an extreme value distribution
P = 1 - e^-E
E = Kmne^-λS
E = mn2^-S′
where S is the bitscore
E < 0.1 considered significant