1. Database Searching:BLAST and FASTA

2. Biological Sequence Databases
NCBI
Nucleotide
GenBank, EMBL, DDBJ, GSDB, patent sequences
Protein
SwissProt, PIR, PRF, PDB (structural)
PopSet
GDB, FlyBase, OMIM, HGMD, MGD, GED
There are other types of biological databases : EcoCyc, WIT/WIT2, others

3. Total amount of non-redundant, well-maintained sequence data is 10- 15 GB and growing expontially
This does not include lots of other easily accessible data like EST’s (expressed sequence tags)

4. In-class exercise: GenBank flatfile format
Open Netscape browser
URL:
Make sure GenBank appears in Search window; type in calmodulin; select Go
Pick any of the returned files; double click to select it
Observe output “flat file”
Remember: the database is not a collection of flatfiles!
You can do this for proteins too, to find the flatfile format for proteins
Locus: unique identifier = accession number
seq length, seq type, some other junk that’s there because of history of database
Definition: summarizes biological sig of record (single text line)
Accession number = primary key for database; stays with the sequence, citable
gi number is geninfo identifier; changes with changes to sequence deposited in DB
lots of other info: keywords, source, organism, reference(s), features (stuff about the sequence, including the sequence!)
Add looking at the tables of GenBank in the server?

5. Database searching: the problem
Experiments yield a sequence, and you want to know what kind of gene/protein it is (or you have a known gene sequence from one organism and you want to know if another organism has a homologous gene)
Databases have millions of sequences
Need method to compare query sequences with all those in databases: speed vs. accuracy

6. The real problem … How do we compare sequences? Seq 1: CTGCACTA
Seq 2: CACTA
or C---ACTA
Add scores

7. The real problem … How do we compare sequences? Seq 1: CTGCACTA
Seq 2: CACTA
or C---ACTA
Scoring tries to approximate evolution: scores for substitutions and for gaps (insertions/deletions)
Scores = sum of terms for substitutions and for gaps (sequence as character string) 41 17

8. Sequence alignment I Simplest scoring: 1 for match, 0 for no match
CTGCACTA
CACTA
C---ACTA
Score = 5
NOTE we are not considering “overhangs” as gaps
Score = 5

9. Sequence alignment II
Slightly more advanced scoring: +1 for match, 0 for no match, -1 for gap
CTGCACTA
CACTA
C---ACTA
Score = 5
again, overhang = endgap is not penalized
Score = 2

10. G C A T G C A T
G C A T G C A T
Identity scoring matrices: top, simple form; below, with mismatch penalty

11. In-class exercise II CCTGGGCTATGC CAGGGTT-TGC
Using the “advanced scoring method” calculate the scores for the following pairs of nucleotide sequences:
CCTGGGCTATGC
CAGGGTT-TGC
CA-GGG-TTTGC
For discussion afterwards: Score you get depends on whether you penalize end gaps. When we get to alignment algorithms, we will discuss this more.

12. Linear vs. affine gap penalties
Linear gap penalty: same penalty subtracted from each space in the gap
Affine gap penalty: first space in the gap has a larger score than subsequent spaces in the gap; i.e., easier to lose/gain more subunits from a gap than to start a new gap/insertion (this makes sense evolutionarily)

13. Linear vs. Affine: example
Match = +1, mismatch = -1, gap = -2;
CCTGGGCTATGC
CC-GG-TT-TGC
Same as above but with affine penalty = -1
CC--GGTT-TGC
Score = 1
Score = 2

14. What about proteins?
Chemistry of amino acids means that some substitutions in the sequence are better than others
Substitution matrix: empirically derived scores for frequency of substitution of each amino acid for all 19 others.
We will talk more about how these substitution matrices are derived later in the course

15. BLOSUM 62 Substitution matrix

16. In-class exercise III
Using the BLOSUM62 substitution matrix and a gap penalty of -2, score the following pairs of protein sequences (do not penalize end gaps)
YIHMNVFLSFML
RVGAANFPNPRL
FIHMNLFVSFML
IHMNLFV--SFML
IVLSMMFFLNHY
These pairs of sequences have the following properties:
First box: unrelated sequences picked arbitrarily from unrelated proteins
Lower box, first column: gaps introduced
Upper box, second column: related, ungapped sequences
Last box: same sequence, randomized

17. In-class exercise IV
Using the BLOSUM62 substitution matrix, find the scores of the following alignment using
A) linear gap penalty of d = 8
B) affine gap penalty of d = 8, e = 2
AVAHV---D--DMPNALS
AAIQLQVTGVVVTDATLK

18. Search algorithms
Early search algorithms scored the whole query sequence against every library sequence; very accurate but very slow
FASTA and BLAST search algorithms use the idea that similar sequences are likely to have small areas of exact matches

19. Original BLAST Algorithm
Make a list of short “words” = wmers from the query sequence:
For nucleotide sequences wmers are used “as is”
for protein sequences wmers are all words which score higher than some threshold T using a substitution matrix
Choose all possible wmers by sliding a window of length w down the query sequence
This is a later algorithm than the FASTA algorithm; we are starting with this one because it is conceptually simpler.

20. W-mers: DNA vs. protein DNA CACTAGCTAAA For w = 6. CACTAG ACTAGC
etc.
Protein
WRKRKKRTGLE
For w=3, T=11
WRK
WRR
WKR
RKR
RKK, etc.

21. Original BLAST algorithm cont.
Scan the library for sequences that match wmers generated in step 1
Extend hits; any extensions must increase the score over that of the original hit; extension stops when no increase in score
wmer
extension
This is illustrated in the movie at (supply address).
Report hit extensions that exceed some threshold score S; the choice of S to include only “similar” sequences is derivable from Karlin Altschul statistics (next week). E is a parameter related to S; it is the expected number of hits returned from a HSP of that length from random matches in the database. Thus a small e-value indicates a low probability of the returned hit being random, which means it has a good probability of being significant.

22. Original BLAST algorithm cont
BLAST results can give more than one area of sequence similarity per pair of proteins compared
BLAST results are more amenable to statistical analysis than FASTA
Add comment about how BLAST works now; reiterate after going through FASTA
Read about BLAST algorithm in text and in the assigned journal paper

23. Gapped BLAST improvements
BLAST now requires two hits within distance A on the same diagonal to trigger extension (overlapping hits are ignored); to keep sensitivity, T is lowered, increasing the number of hits; but since computation time is mostly in the extension phase the overall computation time is much reduced.

24. Gapped BLAST improvement, cont
Gapping is now allowed. This is handled by having a moderate cutoff score for HSP’s that triggers dynamic programming algorithm (discussed in two weeks) to align the two sequences in question
Statistical analysis somewhat compromised, but still very good.

25. In-class exercise IV Open browser; go to www.ncbi.nlm.nih.gov
In “search” window, use pulldown menu to select proteins
Type cytochrome P450 in “for” window; select go
Select a protein by double-clicking
In the database flatfile window, where it says “Default View”, open pulldown menu and select FASTA; then select the Display button
Add link info here.

26. In-class exercise IV, cont.
Paint sequence starting with the first amino acid; select Editcopy;
From top menu bar select Protein; in next window select BLAST from sidebar
In BLAST window, under protein BLAST, select standard protein-protein BLAST
Paste sequence into “search” box; keep the defaults
Select BLAST

27. In-class exercise IV, cont.
In the next window, you will see information about the conserved domain (CD) search that is done by default; you should select Format first, then you can play with the CD results for a while. If you click on the CD image first, you can’t get back and have to start the process over.
The CD search uses profile searching, which we will discuss later in the course
The Format button will open a new window that will display results when the search is over.

28. BLAST results
Bar graph with matches; list of matches with E values (more next time)
Small and intermediate E values are most similar

29. FASTA algorithm AGCTGACGCA CTG GCA
First, look for all identities between small “word” = ktup and every sequence in database. Ktup size determines how many letters must be identical (e.g., 3)
CTGCACTA
CTG
TGC
GCA
etc
AGCTGACGCA
CTG
GCA

30. FASTA, cont a g c t C - T G A
Ktup matches can be depicted in a matrix; diagonals indicate matches. For every library sequence, the 10 best diagonals are constructed from the ktup matches using a distance formula
Read paper again and add stuff notes if necessary.

31. FASTA, cont.
The top 10 diagonals are rescored using substitution matrices and identities smaller than ktup; each of these rescored diagonals is called an initial region.

32. FASTA, cont.
Initial regions are then joined using a joining penalty (like a gap penalty). The highest score of the joined initial regions is then the score for that library sequence. The library sequences are ranked by this score.

33. FASTA, cont.
In the last step of the FASTA algorithm, library sequences that scored above some threshold value are aligned with the query sequence and each other using alignment methods (dynamic programming) we will discuss later.

34. In-class exercise V
Get into seqlab, open bioinfI.list Select FunctionsDatabase Reference SearchingLookup
Select Search the chosen sequence libraries; select GenBank
In the all text box, type hsp70
Select Run; do NOT close the green output window

35. In-class exercise, cont
Making sure nothing in main list is selected, toggle to editor (blank screen)
Select FileAdd Sequences FromDatabases.
Paint the name of the entry GB_BA1:BORHSP70 in the green output window with the left mouse button; click in the Database specification window with the left button, then click with the right button to paste. Select Add to main window.

36. In-class exercise
Close the Database browser window and the green output window
Select the name of the sequence in the editor; select functionsDatabase Sequence SearchingFASTA
Select Search set; then Add database sequences; then Primate under GenEMBL; then click on Add to search set; then close.

37. In-class exercise cont
If there is anything besides the primate library in the search set box, delete it; you can save the primate search set if you want to; then close the search set window.
Select Run on the FASTA window; this will take at least 5 minutes

38. Interpreting FASTA results
FASTA results are reported in a histogram of expected values compared to a random search set. We will discuss this more next time.
The bottom part of the histogram contains the matches of interest (lowest probability of being random)
This explained very well in the text: read this carefully.
Also, pay close attention to the practical suggestions offered in the text for different conditions for using FASTA vs. BLASt

39. Practical considerations in searches
FASTA: varying ktup
BLAST: varying S or E (w optimized by default) in advanced BLAST
Secondary consideration: gap initiation and extension penalties (usual is about 6 and 2 or 6 and 1)
Again, BLAST now allows gapped alignments. This means that the statistical advantage is not quite as clear, though it appears the estimation of the statistical parameters are still v. good.
When to use BLAST and when to use FASTA? supposedly FASTA can be more sensitive on DNA. Many people do not believe this. In practice, if you are doing an exhaustive search for similar sequences you should use both. if you are not, start with BLAST.

40. PSI-BLAST
PSI-BLAST (position-specific-iterated BLAST) is an example of a search method that uses profiles, which we will discuss later in the course
PSI-BLAST starts off with a user-supplied query sequence, and a normal BLAST search is performed

41. If the user decides to iterate (e. g
If the user decides to iterate (e.g., if sequences with good E-values are returned in the first step), then the PSI-BLAST program aligns each returned sequence with the query sequence
The PSI-BLAST program stacks up or aggregates all these individual alignments, producing something that looks like a multiple alignment but is really not

42. From this stack of pairwise alignments, a scoring matrix is calculated by first finding the frequency of amino acids in each column, and then weighting that frequency to take into account very frequent and less frequent amino acids
Then the database is searched again, using this scoring matrix, which takes the variation at each sequence position of the query sequence into account
This procedure can be repeated many times to expand the sequences retrieved by the original query sequence

43. Caveats for PSI-BLAST
The results of a PSI-BLAST search can be skewed greatly by sequences that are recruited into the initial set; “greedy algorithm”
Different but related initial query may give different results (though there is likely to be overlap)
Homework question on PSI-BLAST

44. Other database search methods
SSEARCH: Based on dynamic programming algorithm; we will talk about this in two weeks.
We will talk about PHI-BLAST, BLOCKS, PROBE, BAYES ALIGNER, and other profile-based methods later in the course