A Bit-Parallel, General Integer-Scoring Sequence Alignment Algorithm

Transcription

1 A Bit-Parallel, General Integer-Scoring Sequence Alignment Algorithm GARY BENSON, YOZEN HERNANDEZ, & JOSHUA LOVING B I O I N F O R M A T I C S P R O G R A M B O S T O N U N I V E R S I T Y J L O V I N B U. E D U

2 Introduction: Problem Description Input: Sequences X and Y Integer weights M; I; G match; mismatch; indel or gap that define a similarity or distance scoring function S Output: Calculate the global alignment score for X and Y

3 Sequence Y Introduction Global Alignment Needleman-Wunsch Alignment Scoring Matrix Sequence X A G T C A A

4 Introduction A G T C A A Integer Scores

5 Introduction Penalty from beginning A G T C A A

6 Introduction No initial Penalty A G T C A A

7 Needleman-Wunsch Alignment A -5 G -10 T -15 C -20 A -25 A -30

8 Needleman-Wunsch Alignment A -5 2 G -10 T -15 C -20 A -25 A -30

9 Needleman-Wunsch Alignment A G -10 T -15 C -20 A -25 A -30

10 Needleman-Wunsch Alignment A G -10 T -15 C -20 A -25 A -30

11 Needleman-Wunsch Alignment A G -10 T -15 C -20 A -25 A -30

12 Needleman-Wunsch Alignment A G -10 T -15 C -20 A -25 A -30

13 Needleman-Wunsch Alignment A G -10 T -15 C -20 A -25 A -30

14 Needleman-Wunsch Alignment A G -10 T -15 C -20 A -25 A -30

15 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

16 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

17 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

18 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

19 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

20 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

21 Needleman-Wunsch Alignment A G T -15 C -20 A -25 A -30

22 Bit-parallel alignment A -5 G -10 T -15 C -20 A -25 A -30 Integer Scores

23 Bit-parallel alignment A G -10 T -15 C -20 A -25 A -30 Integer Scores

24 Bit-parallel alignment A G T -15 C -20 A -25 A -30 Integer Scores

25 Bit-parallel alignment A G T C -20 A -25 A -30 Integer Scores

26 Bit-parallel alignment A G T C A -25 A -30 Integer Scores

27 Bit-parallel alignment A G T C A A -30 Integer Scores

28 Bit-parallel alignment A G T C A A Integer Scores

29 Cheaper sequencing of DNA means that larger datasets are being generated Motivation Sequence analysis of such large datasets can be accelerated by faster alignment algorithms Wetterstrand KA. DNA Sequencing Costs: Data from the NHGRI Genome Sequencing Program (GSP) Available at: Accessed June 10, 2013.

30 Earlier Bit-parallel Pattern Matching Algorithms Longest Common Subsequence (LCS) (Allison & Dix, 1986; Crochemore et al, 2001; Hyyro, 2004) Unit-cost edit-distance (Myers, 99; Hyyro et al, 2005) K-differences (agrep; Wu-Manber, 92) Regular expression search (Navarro, 04) Arbitrary weights edit-distance (Bergeron&Hamel, 02)

31 Algorithm Foundation A G T C A A

32 Algorithm Foundation -3-8 A G T C A A

33 Algorithm Foundation A G T C A A

34 Algorithm Foundation H A G V V-21 T NEXT -19 C A A H NEXT

35 Algorithm Foundation Input H H A G V V-21 T V NEXT -19 C A A H NEXT Output

36 Function Table V NEXT output values given V and H input values

37 What is the range of differences?

38 What is the range of differences? Match = 2, Mismatch = -3, Indel = -5

39 What is the range of differences? Match = 2, Mismatch = -3, Indel = -5 Minimum Value = Indel = -5 Maximum Value = Match Indel = 2 - (- 5) = 7

40 Generalized Function Table M = match score I = mismatch score G = indel (gap) penalty

41 Zones in Example Function Table

42 Bit-parallel Representation Bit-vectors: computer words 64 bits long A bit-vector for each possible difference, both horizontally and vertically ( V and H) A set of Match vectors (MatchA, MatchC, MatchG, MatchT in the DNA case) We keep track of match positions because they are a special case in the function table.

43 Example H Bit-vector Storage H values C H Bit-Vectors

44 Example Match Vectors A G T C A A Match Vectors MatchesA MatchesC MatchesT MatchesG

45 Algorithm Start with H values Compute V values Then compute the new H values

46 Algorithm: Example H values C V values A

47 Time Complexity O zn m w where n = Sequence Y m = Sequence X w = length of computer word z = (M 2G + 1)2 (I 2G) 2 2

48 Implementation Python script that generates C code based on input parameters (M; I; G) Will eventually have web page for download of code

49 Experimental Analysis Compared BHL with Wu-Manber K-differences algorithm Unit cost edit distance bit-parallel algorithm Longest Common Subsequence bit-parallel algorithm Needleman-Wunsch dynamic programming algorithm Computed 25 million alignments with each program Each DNA sequence was 63 bases long All programs compiled using GCC, optimization level O3 Computation done on a typical desktop computer

50 Results: Comparison to NW algorithm

51 Results: comparison to bit-parallel algorithms

52 Current and Future Work Implementation for sequences longer than one word Single Instruction Multiple Data (SIMD) implementation BLOSUM and PAM type substitution matrix support General Purpose Graphics Processing Unit (GPGPU) implementation New bit-parallel representations for greater speed and compactness of data

53 Acknowledgements My advisor, Dr. Gary Benson Lab members Yevgeniy Gelfand Yozen Hernandez Funded by the National Science Foundation (NSF)

54 Questions