1. David Wishart David Wishart University of Alberta
February 18th, 2004
BLAST
David Wishart
University of Alberta
Lecture 3.1
(c) 2004 CGDN

2. Objectives
Gain familiarity with sequence searches and comparisons via web-based BLAST
To understand the BLAST algorithm
To understand the principles of BLAST scoring and BLAST statistics
To understand scoring matrices
To become aware of other BLAST services and applications
Lecture 3.1

3. Let’s try an experiment...
ACDEAGHNKLM...
KKDEFGHPKLM...
SCDEFCHLKLM...
Align
MCDEFGHNKLV...
ACDEFGHIKLM...
QCDEFGHAKLM...
AQQQFGHIKLPI...
WCDEFGHLKLM...
SMDEFAHVKLM...
ACDEFGFKKLM...
Lecture 3.1

4. What kind of score distribution do you get?
Lecture 3.1

5. What kind of distribution?
Gaussian?
Poisson?
Other?
Lecture 3.1

6. Extreme Value Distribution
P(x) = e e -e x x
Lecture 3.1

7. Why is this important?
If you can predict the usual score distribution prior to performing an alignment search then it is possible to predict which alignments and which sequences will be worth aligning
Saves on time!
Gives a significance value (not just a raw score) to sequence alignments
Lecture 3.1

8. BLAST Basic Local Alignment Search Tool
Developed in 1990 and 1997 (S. Altschul)
A heuristic method for performing local alignments through searches of high scoring segment pairs (HSP’s)
1st to use statistics to predict significance of initial matches - saves on false leads
Offers both sensitivity and speed
Lecture 3.1

9. BLAST
Looks for clusters of nearby or locally dense “similar or homologous” k-tuples
Uses “look-up” tables to shorten search time
Uses larger “word size” than FASTA to accelerate the search process
Performs both Global and Local alignment
Fastest and most frequently used sequence alignment tool -- THE STANDARD
Lecture 3.1

10. Key References
Altschul, S.F., Gish, W., Miller, W., Myers, E.W. & Lipman, D.J. (1990) "Basic local alignment search tool." J. Mol. Biol. 215:
Altschul, S.F., Madden, T.L., Schäffer, A.A., Zhang, J., Zhang, Z., Miller, W. & Lipman, D.J. (1997)"Gapped BLAST and PSI-BLAST: a new generation of protein database search programs." Nucleic Acids Res. 25:
Lecture 3.1

11. BLAST Access NCBI BLAST Canadian Bioinformatics Resource BLAST
Canadian Bioinformatics Resource BLAST
European Bioinformatics Institute BLAST
Lecture 3.1

12. Lecture 3.1

13. Lecture 3.1

14. Lecture 3.1

15. Different Flavours of BLAST
BLASTP - protein query against protein DB
BLASTN - DNA/RNA query against GenBank (DNA)
BLASTX - 6 frame trans. DNA query against proteinDB
TBLASTN - protein query against 6 frame GB transl.
TBLASTX - 6 frame DNA query to 6 frame GB transl.
PSI-BLAST - protein ‘profile’ query against protein DB
PHI-BLAST - protein pattern against protein DB
Lecture 3.1

16. Other BLAST Services
MEGABLAST - for comparison of large sets of long DNA sequences
RPS-BLAST - Conserved Domain Detection
BLAST 2 Sequences - for performing pairwise alignments for 2 chosen sequences
Genomic BLAST - for alignments against select human, microbial or malarial genomes
VecScreen - for detecting cloning vector contamination in sequenced data
Lecture 3.1

17. Running NCBI BLAST
Lecture 3.1

18. MT0895
MMKIQIYGTGCANCQMLEKNAREAVKELGIDAEFEKIKEMDQILEAGLTALPGLAVDGELKIMGRVASKEEIKKILS
Lecture 3.1

19. Running NCBI BLAST OR OR
Paste in sequence (FASTA format, raw sequence or type in GI or accession number)
>Mysequence MT0895 KIQIYGTGCANCQMLEKNAREAVKELGIDAEFEKIKEMDQILEAGLTALPGLAVDGELKIDS OR
> KIQIYGTGCANCQMLEKNAREAVKELGIDAEFEKIKEMDQILEAGLTALPGLAVDGELKIDS OR
KIQIYGTGCANCQMLEKNAREAVKELGIDAEFEKIKEMDQILEAGLTALPGLAVDGELKIDS
Lecture 3.1

20. Running NCBI BLAST
Choose a range of interest in the sequence “set subsequences” (not usually used)
Select the database from pull-down menu (usually choose nr = non-redundant)
Keep CD Search “check box” on
Leave “Options” unchanged (use defaults)
Go to “Format” menu and adjust Number of descriptions and alignments as desired
Lecture 3.1

21. Running NCBI BLAST
Select Database
Lecture 3.1

22. Conserved Domain Database
Contains a collection of pre-identified functional or structural domains
Derived from Pfam and Smart databases as well as other sources
Uses Reverse Position Specific BLAST (RPS-BLAST) to perform search
Query sequence is compared to a PSSM derived from each of the aligned domains
Lecture 3.1

23. Running NCBI BLAST
Click BLAST!
Lecture 3.1

24. Formatting Results
Lecture 3.1

25. BLAST Format Options
Lecture 3.1

26. BLAST Output
Lecture 3.1

27. BLAST Output
Lecture 3.1

28. BLAST Output
Lecture 3.1

29. BLAST Output
Lecture 3.1

30. BLAST Output
Lecture 3.1

31. BLAST Output
Lecture 3.1

32. BLAST Parameters Identities - No. & % exact residue matches
Positives - No. and % similar & ID matches
Gaps - No. & % gaps introduced
Score - Summed HSP score (S)
Bit Score - a normalized score (S’)
Expect (E) - Expected # of chance HSP aligns
P - Probability of getting a score > X
T - Minimum word or k-tuple score (Threshold)
Lecture 3.1

33. BLAST - Rules of Thumb
Expect (E-value) is equal to the number of BLAST alignments with a given Score that are expected to be seen simply due to chance
Don’t trust a BLAST alignment with an Expect score > 0.01 (Grey zone is between )
Expect and Score are related, but Expect contains more information. Note that %Identies is more useful than the bit Score
Recall Doolittle’s Curve (%ID vs. Length, next slide) %ID > 30 - numres/50
If uncertain about a hit, perform a PSI-BLAST search
Lecture 3.1

34. Doolittle’s Curve
Twilight Zone
Lecture 3.1

35. BLAST Statistics
Lecture 3.1

36. Extreme Value Distribution
Lecture 3.1

37. Extreme Value Distribution
Kmne-lS is called Expect or E-value
In BLAST E = 10 so P =
If E is small (<0.01) then P is small
If Matches = 1 and Mismatches = -1 then:
l = lnq/p and K = (q-p)2/q
p = probability of match = 0.05
q = probability of not match = 0.95
Then l = 2.94 and K =0.85
m = length of sequence & n = length of database
S = score for given HSP
Lecture 3.1

38. How Does BLAST Really Work?
Lecture 3.1

39. BLAST Algorithm Query: TPQGQRQGQ….. TPQ PQG QGQ GQR QRQ RQG …,
AAA AAC AAD PQG QGQ YYY
AGA AGC AAN … PEG QGM
AAG GAC AAE … PRG MGQ
GAA AAQ … PMG QAQ
GAG QGN
Lecture 3.1

40. BLAST Algorithm Query: TPQGQRQGQ….. AAA AAC AAD ... PQG QGQ YYY
AGA AGC AAN … PEG QGM
AAG GAC AAE … PRG MGQ
GAA AAQ … PMG QAQ
Database: CTVTPMGQREAE…
HSP
Lecture 3.1

41. High-scoring Segment Pairs
PQG 18
PEG 15
PRG 14
PMG 14
PNG 13
PDG 13
PQG 12
etc.
Query: LNKCKTPQGQRQGQQWIKQPLMDKN
L TP+GQR++++W+ P+ D
Sbjct: LDCTVTPMGQREAERWLHMPVRDTR T
Lecture 3.1

42. Extending HSP’s E = kNe X Cumulative Score S T Extension (# aa) -ls
Number of HSP’s found
purely by chance
-ls X T S
Lecture 3.1

43. Visualizing HSP’s
T P Q G Q R Q G Q C T V
P M G Q R
Lecture 3.1

44. Connecting HSP’s
Lecture 3.1

45. The Final Result
Lecture 3.1

46. Getting the Most from BLAST
Lecture 3.1

47. BLAST Options
Lecture 3.1

48. BLAST Options Composition-based statistics (Yes)
Sequence Complexity Filter (Yes)
Expect (E) value (10)
Word Size (3)
Substitution or Scoring Matrix (Blosum62)
Gap Insertion Penalty (11)
Gap Extension Penalty (1)
Lecture 3.1

49. Composition Statistics
Recent addition to BLAST algorithm
Permits calculated E (Expect) values to account for amino acid composition of queries and database hits
Improves accuracy and reduces false positives
Effectively conducts a different scoring procedure for each sequence in database
Lecture 3.1

50. LCR’s (low complexity)
Watch out for…
transmembrane or signal peptide regions
coil-coil regions
short amino acid repeats (collagen, elastin)
homopolymeric repeats
BLAST uses SEG to mask amino acids
BLAST uses DUST to mask bases
Lecture 3.1

51. Scoring Matrices BLOSUM Matrices PAM Matrices
Developed by Henikoff & Henikoff (1992)
BLOcks SUbstitution Matrix
Derived from the BLOCKS database
PAM Matrices
Developed by Schwarz and Dayhoff (1978)
Point Accepted Mutation
Derived from manual alignments of closely related proteins
Lecture 3.1

52. How to Make Your Own Matrix
C D ...
ACDEFGH..
ACDEFGK..
AADEFGH..
GCDEFGH..
ACAEYGK..
ACAEFAH..
#Aobs
f(A,A) = A
#Aexp C D
#C/Aobs
f(C,A) = E
#Aexp +
#Cexp
Perform Calculate Fill Sub
Alignment Frequencies Matrix
Lecture 3.1

53. How to Make a PAM Matrix X = Multiply Matrices N times to make PAM “X”
log
Take the log
Lecture 3.1

54. PAM versus BLOSUM First useful scoring matrix for protein
Assumed a Markov Model of evolution (I.e. all sites equally mutable and independent)
Derived from small, closely related proteins with ~15% divergence
Much later entry to matrix “sweepstakes”
No evolutionary model is assumed
Built from PROSITE derived sequence blocks
Uses much larger, more diverse set of protein sequences (30% - 90% ID)
Lecture 3.1

55. PAM versus BLOSUM
Higher PAM numbers to detect more remote sequence similarities
Lower PAM numbers to detect high similarities
1 PAM ~ 1 million years of divergence
Errors in PAM 1 are scaled 250X in PAM 250
Lower BLOSUM numbers to detect more remote sequence similarities
Higher BLOSUM numbers to detect high similarities
Sensitive to structural and functional subsitution
Errors in BLOSUM arise from errors in alignment
Lecture 3.1

56. PAM Matricies
PAM 40 - prepared by multiplying PAM 1 by itself a total of 40 times best for short alignments with high similarity
PAM prepared by multiplying PAM 1 by itself a total of 120 times best for general alignment
PAM prepared by multiplying PAM 1 by itself a total of 250 times best for detecting distant sequence similarity
Lecture 3.1

57. PAM250
Lecture 3.1

58. BLOSUM Matricies
BLOSUM 90 - prepared from BLOCKS sequences with >90% sequence ID best for short alignments with high similarity
BLOSUM 62 - prepared from BLOCKS sequences with >62% sequence ID best for general alignment (default)
BLOSUM 30 - prepared from BLOCKS sequences with >30% sequence ID best for detecting weak local alignments
Lecture 3.1

59. BLOSUM62
Lecture 3.1

60. Scraping the Bottom of the Barrel with Psi-BLAST
Lecture 3.1

61. PSI-BLAST Algorithm
Perform initial alignment with BLAST using BLOSUM 62 substitution matrix
Construct a multiple alignment from matches
Prepare position specific scoring matrix
Use PSSM profile as the scoring matrix for a second BLAST run against database
Repeat steps 3-5 until convergence
Lecture 3.1

62. Position Specific Scoring Matrix (PSSM)
<e>i = log2(qi/pi)
Lecture 3.1

63. PSI-BLAST
Lecture 3.1

64. PSI-BLAST
PresS Iterate!
Lecture 3.1

65. PSI-BLAST
PresS Iterate!
Lecture 3.1

66. PSI-BLAST
Lecture 3.1

67. PSI-BLAST For Protein Sequences ONLY Much more sensitive than BLAST
Slower (iterative process)
Often yields results that are as good as many common threading methods
SHOULD BE YOUR FIRST CHOICE IN ANALYZING A NEW SEQUENCE
Lecture 3.1

68. BLAST against PDB
Lecture 3.1

69. Still Confused?
Lecture 3.1

70. Conclusions
BLAST is the most important program in bioinformatics (maybe all of biology)
BLAST is based on sound statistical principles (key to its speed and sensitivity)
A basic understanding of its principles is key for using/interpreting BLAST output
Use NBLAST or MEGABLAST for DNA
Use PSI-BLAST for protein searches
Lecture 3.1