Review of BLAST

1
Review of BLAST

• Heuristic, local alignment algorithm
• Permits trade-off between speed and sensitivity
• HSPs are dependent on matrix used
• Statistical significance of results can be
  calculated by E-Value

significant matches detected
sensitivity
__________________________________________________
________________________
significant matches in DB
2
Refinements of BLAST

• Two-hit method
• Gapped Alignments
• Position-Specific Iteration
• Gapped BLAST and PSI-BLAST a new generation of
  protein database search programs SF Altschul, TL
  Madden, AA Schaffer, J Zhang, Z Zhang, W Miller,
  and DJ Lipman Nucl. Acids Res. 25 3389-3402.
  http//nar.oxfordjournals.org/cgi/content/full/25/
  17/3389

3
Two-hit method

• BLAST v1
• Seeks short word pairs with aligned score T
• Each hit is extended to test if it is within a
  high-scoring alignment (consumes most processing
  time)
• BLAST v2
• Seeks two non-overlapping word pairs on the same
  diagonal, within a certain distance
• T is lowered yielding more hits
• Fewer number of two non-overlapping word pairs
  exist decreasing average compute time

4
Gapped Alignments

• BLAST v1
• Finds several alignments involving a single
  database sequence
• When alignments are combined, resulting alignment
  is statistically significant
• When the alignments are not combined, individual
  alignments may not meet statistical threshold to
  be reported
• BLAST v2
• Introduces an algorithm to generate gapped
  alignments overcoming issues with BLAST v1
• Allows T to be raised increasing speed of initial
  database scan
• Gapped alignment algorithm uses DP to extend a
  central pair of aligned residues in both
  directions confined to a pre-defined strip of the
  DP path graph

5
PSI-BLAST Position-Specific Iterative BLAST

• Motif or profile search methods are much more
  sensitive than pairwise comparison methods at
  detecting distant relationships
• Basic Idea
• BLAST searches may be iterated, with a
  position-specific score matrix generated from
  significant alignments found in round i used for
  round i 1

6
Definition of Profile/Motif

• an analysis representing the extent to which
  something exhibits various characteristics
• Examples from ProSite
• N-glycosylation site
• N-P-ST-P
• Glycosaminoglycan attachment site
• S-G-x-G
• cAMP- and cGMP-dependent protein kinase
  phosphorylation site
• RK(2)-x-ST

7
Position-Specific Scoring Matrix

• A PSSM is a motif descriptor
• The description includes a weight (score,
  probability, likelihood) for each symbol
  occurring at each position along the motif
• Examples of motifs
• Protein active sites, structural elements, zinc
  finger, intron/exon boundaries,
  transcription-factor binding sites, etc.

8
PSSM Example

• For DNA
• GTA-AT-G-AC-N-TAC

9
PSSM Example

• For DNA
• GTA-AT-G-AC-N-TAC

10
Position-Specific Scoring Matrix

• Construction of PSSM is a multi-stage process
• Architecture of matrix
• Create multiple alignment from which the matrix
  is derived
• Calculate frequencies for each position
• Applying BLAST to PSSM

11
Position-Specific Scoring Matrix

• 10 vertebrate donor site sequences aligned at
  exon/intron boundary

12
Position-Specific Scoring Matrix

• Calculate the absolute frequency of each
  nucleotide at each position

13
Position-Specific Scoring Matrix

• Calculate the absolute frequency of each
  nucleotide at each position

14
Position-Specific Scoring Matrix

• Calculate the relative frequency of each
  nucleotide at each position

15
Position-Specific Scoring Matrix

• Calculate the relative frequency of each
  nucleotide at each position

16
Position-Specific Scoring Matrix

• What is the probability of finding CAGGTTGGA?
• The product of the frequency of each nucleotide
  at each position
• C is 0.2 at position 1, A is 0.6 at position 2,
  etc -gt 0.2 0.6 0.7 1 1 0.1 0.1 0.5
  0.1

17
Position-Specific Scoring Matrix

• The ratio of the probability of a sequence in a
  given model, P(SM), and the probability of a
  sequence in a random model, P(SR) is a
  likelihood ratio (or odds ratio) P(SM) / P(SR)
• And its logarithm is a log likelihood ratio (or
  log odds ratio)
• If the ratio is
• 0, S has the same probability to appear in M as
  in R
• gt0, S is more likely to appear in M than in R
• lt 0, S is less likely to appear in M than in R

18
Position-Specific Scoring Matrix

• Compute the log-likelihood values with the
  transformation ln(Mij/Pi)
• where Mij is the probability of nucleotide i at
  position j and Pi is background probability of of
  nucleotide i
• Assume each nucleotide can appear at any
  position, then Pi0.25 in random model

19
Position-Specific Scoring Matrix

• Compute the log-likelihood values with the
  transformation ln(Mij/Pi)
• where Mij is the probability of nucleotide i at
  position j and Pi is background probability of of
  nucleotide i
• Assume each nucleotide can appear at any
  position, then Pi0.25 in random model

20
Position-Specific Scoring Matrix

• Use the profile to scan a sequence
• Sum the coefficients from the matrix for each
  nucleotide in each position
• Formally, matrix M for a sequence s of length l
  (s s1, ... , sl, and sk being one of A, C, G,
  T) is computed as

21
Position-Specific Scoring Matrix

• Assume the sequence GTAGTAGAAGGTAAGTGTCCGTAG
• Find the score forGTAGTAgaaggtaagTGTCCGTAGGTAGTA
  GaaggtaagtGTCCGTAGGTAGTAGAAGGtaagtgtccGTAG

22
Position-Specific Scoring Matrix

• Assume the sequence GTAGTAGAAGGTAAGTGTCCGTAG
• Find the score forGTAGTAgaaggtaagTGTCCGTAG
  (-8)GTAGTAGaaggtaagtGTCCGTAGGTAGTAGAAGGtaagtgtcc
  GTAG

23
Position-Specific Scoring Matrix

• Assume the sequence GTAGTAGAAGGTAAGTGTCCGTAG
• Find the score forGTAGTAgaaggtaagTGTCCGTAG
  (-8)GTAGTAGaaggtaagtGTCCGTAG (8.33)GTAGTAGAAGGta
  agtgtccGTAG

24
Position-Specific Scoring Matrix

• Assume the sequence GTAGTAGAAGGTAAGTGTCCGTAG
• Find the score forGTAGTAgaaggtaagTGTCCGTAG
  (-8)GTAGTAGaaggtaagtGTCCGTAG (8.33)GTAGTAGAAGGta
  agtgtccGTAG (0.24)

25
PSI-BLAST

• Confirming relationships of purine
• nucleotide metabolism proteins

26
PSI-BLAST
e value cutoff for PSSM
27
RESULTS Initial BLASTP
Same results as protein-protein BLAST
28
Results of First PSSM Search
Other purine nucleotide metabolizing enzymes not
found by ordinary BLAST
29
Third PSSM Search Convergence