PatternHunter II: Highly Sensitive and Fast Homology Search

1
PatternHunter II Highly Sensitive and Fast
Homology Search
Bioinformatics and Computational Molecular
Biology (Fall 2005) Representation
Ming Li, Bin Ma Derek Kisman, John Tromp
R94922059 ???
2
Overview

• Homology search
• Local alignment algorithms
• PH I
• PH II
• Multiple Spaced Seeds
• Computing hit probability
• Finding a good seed set
• PH II Design
• Performance

3
Local alignment

• Smith-Waterman
• Smith and Waterman, 1981 Waterman and Eggert,
  1987
• SSearch
• FastA
• Wilbur and Lipman, 1983 Lipman and Pearson, 1985
• BLAST
• Altschul et al., 1990 Altschul et al., 1997
• Blast Family BLASTN, BLASTP, etc.
• MEGABLAST

4
PatternHunter

• Seed
• Tradeoff sensitivity lt-gt computation
• Consecutive k letters
• k11 in Blastn, k28 in MegaBlast
• Nonconsecutive k letters
• Spaced seed
• A model of k as its weight

5
PatternHunter II

• Genome Informatics 14 (2003)
• Extend single optimized spaced seed of PH to
  multiple ones
• Speed BLASTN (MEGABLAST)
• Sensitivity Smith-Waterman (SSearch)

6
Definition

• A homologous region, R
• A seed hits R
• A seed set Aa1,ak hits R
• Similarity
• R has px identities
• Sensitivity
• Hit probability
• Optimal (DP) 1

7
Computing Hit Probability

• NP-hard on multiple seeds
• DP on 1 seed
• Extend DP to multiple seeds

8
Computing Hit Probability of Multiple Seeds

• Let Aa1,ak be a set of k seeds and R a random
  region of Length L with similarity level p.
• Binary string b is a suffix of R0i
• Answer f ( L,? ), ? empty string

9
Computing Hit Probability of Multiple Seeds
10
Computing Hit Probability of Multiple Seeds
11
Finding a Good Seed Set

• NP-hard for both optimal seed and multiple seeds
• Greedy

12
Finding a Good Seed Set

• Compute the 1st seed a1 which maximizes the hit
  probability of a1
• Compute the 2nd seed a2 which maximizes the hit
  probability of a1, a2
• Repeat until
• Reach the desired number of seeds
• Reach the desired hit probability

13
Finding a Good Seed Set

• May not optimize the combined hit probability
• Good enough
• Optimal
• 16 weight, 11 seeds, L64, similarity70, first
  four seeds111010010100110111,1111001100101000010
  11,110100001100010101111,1110111010001111
• Greedy
• 16 weight, 12 seeds, L64, similarity70, first
  four seeds111010010100110111,1111000100010011010
  111,1100110100101000110111,1110100011110010001101
  

14
Performance of the seeds

• From low to high
• Solid weight-11 k1,2,4,8,16 seeds
• Dashed 1-seed, weight10,9,8,7

15
Performance of the seeds

• Reducing the weight by 1
• Increase the expected number of hits by a factor
  of 4
• Doubling the number of seeds
• Increase the expected number of hits by a factor
  of 2
• Better Multiple seeds

16
PH II Performance

• Compare with Blast(Blastn), Smith-Waterman(SSearch
  )
• Sensitivity of SSearch 1
• Alignment score
• BLAST methods (hash, DP)
• match1, mismatch-1, gapopen-5, gapextension-1

17
PH II Performance

• From low to high
• Solid PH II, 1, 2, 4, 8 seeds weight 11
• Dashed Blastn, seed weight 11

18
Complexity Proof

• Finding optimal spaced seeds
• NP-hard
• Finding one optimal seed
• NP-hard
• Computing the hit probability of multiple seeds
• NP-hard