Protein Multiple Alignment

1
Protein Multiple Alignment

• by Konstantin Davydov

2
Papers

• MUSCLE a multiple sequence alignment method with
  reduced time and space complexity by Robert C
  Edgar
• ProbCons Probabilistic Consistency-based
  Multiple Sequence Alignment by Chuong B. Do,
  Mahathi S. P. Mahabhashyam, Michael Brudno, and
  Serafim Batzoglou

3
Outline

• Introduction
• Background
• MUSCLE
• ProbCons
• Conclusion

4
Introduction

• What is multiple protein alignment?
• Given N sequences of amino acids x1, x2 xN
• Insert gaps in each of the xis so that
• All sequences have the same length
• Score of the global map is maximum

5
Introduction

• Motivation
• Phylogenetic tree estimation
• Secondary structure prediction
• Identification of critical regions

6
Outline

• Introduction
• Background
• MUSCLE
• ProbCons
• Conclusion

7
Background

• Aligning two sequences

8
Background
AGTGCCCTGGAACCCTGACGGTGGGTCACAAAACTTCTGGA
AGTGACCTGGGAAGACCCTGACCCTGGGTCACAAAACTC
9
Background

• Unfortunately, this can get very expensive
• Aligning N sequences of length L requires a
  matrix of size LN, where each square in the
  matrix has 2N-1 neighbors
• This gives a total time complexity of
• O(2N LN)

10
Outline

• Introduction
• Background
• MUSCLE
• ProbCons
• Conclusion

11
MUSCLE
12
MUSCLE

• Basic Strategy A progressive alignment is built,
  to which horizontal refinement is applied
• Three stages
• At end of each stage, a multiple alignment is
  available and the algorithm can be terminated

13
Three Stages

• Draft Progressive
• Improved Progressive
• Refinement

14
Stage 1 Draft Progressive

• Similarity Measure
• Calculated using k-mer counting.

ACCATGCGAATGGTCCACAATG
k-mer ATG CCA
score 3
2
15
Stage 1 Draft Progressive

• Distance estimate
• Based on the similarities, construct a triangular
  distance matrix.

16
Stage 1 Draft Progressive

• Tree construction
• From the distance matrix we construct a tree

17
Stage 1 Draft Progressive

18
Stage 1 Draft Progressive

• Progressive alignment
• A progressive alignment is built by following the
  branching order of the tree. This yields a
  multiple alignment of all input sequences at the
  root.

19
Stage 2 Improved Progressive

• Attempts to improve the tree and uses it to build
  a new progressive alignment. This stage may be
  iterated.

20
Stage 2 Improved Progressive

• Similarity Measure
• Similarity is calculated for each pair of
  sequences using fractional identity computed from
  their mutual alignment in the current multiple
  alignment

21
Stage 2 Improved Progressive

• Tree construction
• A tree is constructed by computing a Kimura
  distance matrix and applying a clustering method
  to it

22
Stage 2 Improved Progressive

• Tree comparison
• The new tree is compared to the previous tree by
  identifying the set of internal nodes for which
  the branching order has changed

23
Stage 2 Improved Progressive

• Progressive alignment
• A new progressive alignment is built

24
Stage 3 Refinement

• Performs iterative refinement

25
Stage 3 Refinement

• Choice of bipartition
• An edge is removed from the tree, dividing the
  sequences into two disjoint subsets

X1
X3
X2
X4
X5
26
Stage 3 Refinement

• Profile Extraction
• The multiple alignment of each subset is
  extracted from current multiple alignment.
  Columns made up of indels only are removed

27
Stage 3 Refinement

• Re-alignment
• The two profiles are then realigned with each
  other using profile-profile alignment.

28
Stage 3 Refinement

• Accept/Reject
• The score of the new alignment is computed, if
  the score is higher than the old alignment, the
  new alignment is retained, otherwise it is
  discarded.

29
MUSCLE Review

• Performance
• For alignment of N sequences of length L
• Space complexity O(N2L2)
• Time complexity O(N4NL2)
• Time complexity without refinement O(N3NL2)

30
Outline

• Introduction
• Background
• MUSCLE
• ProbCons
• Conclusion

31
Hidden Markov Models (HMMs)
X
M
X
AGCC-AGC
Y
-GCCCAGT
Y
IMMMJMMM
J
I
--
--
X
Y
32
Pairwise Alignment

• Viterbi Algorithm
• Picks the alignment that is most likely to be the
  optimal alignment
• However the most likely alignment is not the
  most accurate
• Alternative find the alignment of maximum
  expected accuracy

33
Lazy Teacher Analogy

• 10 students take a 10-question true-false quiz
• How do you make the answer key?
• Viterbi Approach Use the answer sheet of the
  best student
• MEA Approach Weighted majority vote

34
Viterbi vs MEA

• Viterbi
• Picks the alignment with the highest chance of
  being completely correct
• Maximum Expected Accuracy
• Picks the alignment with the highest expected
  number of correct predictions

35
ProbCons

• Basic Strategy Uses Hidden Markov Models (HMM)
  to predict the probability of an alignment.
• Uses Maximum Expected Accuracy instead of the
  Viterbi alignment.
• 5 steps

36
Notation

• Given N sequences
• S s1, s2, sN
• a is the optimal alignment

37
ProbCons

• Step 1 Computation of posterior-probability
  matrices
• Step 2 Computation of expected accuracies
• Step 3 Probabilistic consistency transformation
• Step 4 Computation of guide tree
• Step 5 Progressive alignment
• Post-processing step Iterative refinement

38
Step 1 Computation of posterior-probability
matrices

• For every pair of sequences x,y?S, compute the
  matrix Pxy
• Pxy(i, j) P(xiyj ?a x, y), which is the
  probability that xi and yj are paired in a

39
Step 2 Computation of expected accuracies

• For a pairwise alignment a between x and y,
  define the accuracy as

40
Step 2 Computation of expected accuracies
(continued)

• MEA alignment is found by finding the highest
  summing path through the matrix
• Mxyi, j P(xi is aligned to yj x, y)

41
Consistency
zk
z
xi
x
y
yj
yj
42
Step 3 Probabilistic consistency transformation

• Re-estimate the match quality scores P(xiyj ?a
  x, y) by applying the probabilistic consistency
  transformation which incorporates similarity of x
  and y to other sequences from S into the x-y
  comparison

P(xiyj ?a x, y)
P(xiyj ?a x, y, z)
43
Step 3 Probabilistic consistency transformation
(continued)
44
Step 3 Probabilistic consistency transformation
(continued)

• Since most of the values of Pxz and Pzy will be
  very small, we ignore all the entries in which
  the value is smaller than some threshold w.
• Use sparse matrix multiplication
• May be repeated

45
Step 4 Computation of guide tree

• Use E(x,y) as a measure of similarity
• Define similarity of two clusters by the
  sum-of-pairs

46
Step 5 Progressive alignment

• Align sequence groups hierarchically according to
  the order specified in the guide tree.
• Alignments are scored using sum-of-pairs scoring
  function.
• Aligned residues are scored according to the
  match quality scores P(xiyj ?a x, y)
• Gap penalties are set to 0.

47
Post-processing step iterative refinement

• Much like in MUSCLE
• Randomly partition alignment into two groups of
  sequences and realign
• May be repeated

48
ProbCons overview

• ProbCons demonstrated dramatic improvements in
  alignment accuracy
• Longer running time
• Doesnt use protein-specific alignment
  information, so can be used to align DNA
  sequences with improved accuracy over the
  Needleman-Wunsch algorithm.

49
Outline

• Introduction
• Background
• MUSCLE
• ProbCons
• Conclusion

50
Conclusion

• MUSCLE demonstrated poor accuracy, but very short
  running time.
• ProbCons demonstrated dramatic improvements in
  alignment accuracy, however, is much slower than
  MUSCLE.

51
Results
52
Reliability Scores
53
Questions?
54
References

• Robert C Edgar
• MUSCLE a multiple sequence alignment method with
  reduced time and space complexity
• Chuong B. Do, Mahathi S. P. Mahabhashyam, Michael
  Brudno, and Serafim Batzoglou
• ProbCons Probabilistic Consistency-based
  Multiple Sequence Alignment

55
References (continued)

• Slides on Multiple Sequence Alignment, CS262
• Slides on Sequence similarity, CS273
• Slides on Protein Multiple Alignment, Marina
  Sirota