Parallel DNA Alignment

1
Parallel DNA Alignment

• Jiang Lu

2
DNA Pairwise Alignment

• DNA stores genetic information
• by the permutation of the four
• nucleotide A, C, G, T.
• Global alignment
• For two similar sequence
• End to end
• Local alignment
• Not similar
• The best matched part

Global CTCTACGCCAGAG C- -TAC -CCA
-AG Local CTCTACGCC- AGAG - - CTAC-
CCAAG
3
Algorithm Development Timeline

DP Optimal but slow Seed Fast but no guarantee
4
Dynamic Programming (global alignment)
X1
X2
X3

Xi-1
Xi

• If xi yj then the aligment is xi add
  alignment of Xi -1 and Yj -1.
• If xi ? yj then the alignment is the alignment
  of Xi -1 and Yj
• If xi ? yj then the alignment is the alignment
  of Xi and Yj -1.
• Demo http//baba.sourceforge.net/

Y1
Y2
Y3

Yj-1
Yj
5
Fine-grained parallel based on SIMD

• Parallel calculate multiple lines/columns/blocks
• Front-wave parallel compute diagonally

6
BLASTThe most widely used tool for alignment

• Good performance 100 times faster than dynamic
  programming
• Demo http//blast.ncbi.nlm.nih.gov

7
Ordered Index Seed Algorithm (I) sequence
indexing

• A new seed based algorithm published in 2008 IEEE
• Its claimed 528 times faster than BLAST
• First step encode and indexing

8
Ordered Index Seed Algorithm (II) Un-gapped
extension

• For each seed, match the neighbor pair in two
  directions
• For seed AACTGTAA (using 8-length seed)
• Skip duplicate extension, e.g. seed AATTGCTC
• Score each extension and keep the one higher than
  the threshold
• Match 5 points
• Unmatch - 4 points

9
Ordered Index Seed Algorithm (III) Gapped
extension

• In mutation process, the
• gene sequence maybe
• interrupted
• Use dynamic programming
• do gapped extension in
• middle of the seed

10
Coarse-grained parallel application IBM's gene
database on Blue Gene/L

• Process 2 million alignment against 2.5 million
  sequences per day. - The fastest when published
  at 2005

11
Performance

12
Question Sheet

• What are the two DNA alignment types?
• What are the two main algorithm categories of
  pairwise DNA alignment?
• What is the most widely used DNA alignment
  algorithm?