Aligning Multiple Genome Sequences With the Threaded Blockset Aligner

1
Aligning Multiple Genome Sequences With the
Threaded Blockset Aligner

• Blanchette, W., Kent, W.J., Riemer, C., Elnitski,
  L., Smit, A.F.A., Roskin, K.M., Baertsch, R.,
  Rosenbloom, K., Clawson, H., Green, E.D.,
  Haussler, D., and Miller, W.
• Genome Research 2004

2
Outline

• Introduction
• TBA
• MULTIZ
• How TBA was built
• Evaluation of alignment accuracy
• Accuracy of the Multiple Alignments
• Experiment results

3
Introduction

• Reference Sequence Idea
• A sequence is fixed as the reference to which all
  other sequences are compared

S1 A T G C T C S2 A G A G C S3 T T C T
G S4 A T T G C A T G C
S1 D(S1,S2) D(S1,S3) D(S1,S4) 9 S2
D(S2,S1) D(S2,S3) D(S2,S4) 12 S3 D(S3,S1)
D(S3,S2) D(S3,S4) 12 S4 D(S4,S1)
D(S4,S2) D(S4,S3) 11
S1 A T - G C - T - C S2 A - -
G A - G - C S3 - T - T C - T -
G S4 A T T G C A T G C
S1 A T G C T C S2 A - G A
G C
S1 A T G C T C S2 A - G A
G C S3 - T T C T G
Efficient methods for multiple sequence alignment
with guaranteed error bounds, Gusfield, D., Bull.
Math. Biol., 1993, Vol. 55, pp. 141-54.
4

• Benefit
• Simplicity
• Drawbacks
• Regions conserved in a subset of the species, but
  absent from the reference sequence, are not
  identified.
• Alignments generated with different reference
  sequences may be inconsistent.
• Inconsistent
• Two positions that are aligned to each other
  using one reference sequence might be aligned to
  different positions when another reference
  sequence is chosen.

S1 A T - G C - T - C S2 A - -
G A - G - C S3 - T - T C - T -
G S4 A T T G C A T G C
S1 A T G C T C S2 A G A G C S3 T T C T
G S4 A T T G C A T G C
5
TBA

• Threaded Blockset Aligner
• Block
• A local alignment of the sequences
• Blockset
• A set of Blocks

6
TBA
h human (400bp) m mouse (400bp) r rat
(350bp)
7
TBA

• Thread
• A sequence S threads a blockset if every position
  in the sequence S appears exactly once in some
  block of the blockset.
• Threaded blockset
• A blockset is threaded by each of the original
  sequences.

8
TBA
h human (400bp) m mouse (400bp) r rat
(350bp)
9

• Ref-blockset
• A Blockset where every block has a row from a
  particular sequence which is designated as the
  reference for that ref-blockset.
• Projection
• Given a thread blockset,
• generate an S-ref blockset for any sequence S.

10
TBA

• Any two ref-blocksets generated by projection
  from the same threaded blockset are consistent.

11
TBA

• Threaded Blockset Aligner
• TBA produces a set of blocks in which each
  position in the given sequences to be aligned
  appears once and only once.
• Any detected match among some or all of the
  sequences is represented among the blocks, and
  mutually consistent reference-sequence alignments
  can be extracted at will.

12

• Alignment between the chloroplast genomes of
  Arabidopsis thaliana(???? ) and Oenothera
  elata(??? ) by PipMaker.
• Blocks of a threaded blockset for the chloroplast
  genomes of Arabidopsis thaliana(a) and Oenothera
  elata(p).

13
Applying TBA to vertebrate HOX clusters
Tilapia
Mammals
Fish
14
Applying TBA to vertebrate HOX clusters
Human
Mammals
Fish
15

• Assumption
• The matching regions occur in the same order and
  orientation in all species.
• Partial order
• For a sequence S, Ss segments in block A
  precedes Ss segments in block B, and we say that
  block A precedes block B.
• Local alignment
• Pairwise alignment BLATZ
• Three or more sequences alignments MULITZ

16
MULTIZ

• Deals with alignments between three or more
  sequences .
• MULTIZ
• Merge two blocksets by assistance of another
  guiding blockset.
• HUMOR
• A specialized version of MULTIZ used in The Rat
  Genome Sequencing Consortium 2003.s

17
How does it work?
18
How does it work? Cont.

• Proceeds in order along S (The reference for G, M
  and the output).
• Access the corresponding (to Ss position)
  portion of N according to G.
• Collect each aligned columns.

19
HUMOR

• Stands for Human-Mouse-Rat
• Starts with pairwise human-ref blocksets for
  human-mouse and for human-rat.
• Trims columns from the ends of the blocks to make
  the human components identical.
• Aligns the mouse and rat intervals to each
  other.
• Aligns the human interval to the resulting
  mouse-rat block.

20
How TBA was built
21
(No Transcript)
22
Evaluation of Alignment Accuracy

• Simulate sequence evolution, starting with some
  ancestral sequence and performing mutation along
  the branches of a predetermined phylogenetic
  tree.
• Use the agreement between the truth and the
  result as a scoring method.

23
Accuracy of the Multiple Alignments (9 Mammals)
24
Accuracy of the Multiple Alignments (H,M,R)
25
Experimental results

• Accuracy of the closely related sequences is
  better than more diverged ones.
• TBA uniformly stands out for the more diverged
  pairs.
• For most programs, their accuracy increases when
  therere smaller number of species (indicates
  improvement, more species should have more
  information).

26
Experimental results

• MULTIZ suffers mouse-rat alignment.
• Human-rat is also slightly worse than the
  human-mouse alignment because rat is aligned to
  human only through mouse.
• Score of 1.0 may be impossible to achieve,
  because a certain information is lost during
  sequence evolution.
• Score of 1.0 is usually not necessary, some
  errors are inconsequential.

27
Experimental results

• Running Time
• Only the four programs (MULTIZ, TBA, MAVID,
  MLAGAN) actually designed for aligning large
  regions run fast enough.
• MAVID super fast!

28

• Thank you

29
(No Transcript)