Bioinformatics

1
Lecture 9
Bioinformatics

• Multiple sequence alignments
• Scoring multiple sequence alignments
• Progressive methods
• ClustalW
• Other methods
• Hidden Markov Models

2
Multiple sequence alignments (MSA)

• Comparing a pair of sequences is not sufficient
  for many research purposes, mainly for
  evolutionary reconstructions and study functional
  similarities.
• It is obvious that MSA is much more demanding in
  computational sense. For two protein sequences
  each 300 aa in length and excluding gaps, the
  number of comparisons to be made using dynamic
  programming approach is equal to 3002 9 x 104.
  For 3 sequences of the same length this number is
  3003 2.7 x 107. For 10 sequences it becomes
  staggering.
• Fortunately in late 1980 and mid 1990 methods,
  which dramatically reduce a number of
  comparisons, were invented.
• The MSA alignment is usually done in three
  consecutive steps. 1. Finding alignments between
  each pair of sequences 2. A trial MSA is then
  produced by predicting a phylogenetic tree for
  the sequences (for instance neighbor-joining
  method) 3. The sequences are then multiply
  aligned in the order of their relationship on the
  tree.

3
Scoring Multiple sequence alignments
4
Progressive methods of MSA

• The most closely related sequences are first
  aligned by dynamic programming to build a MSA
  starting from the most related sequences
• The tree is based on pairwise comparisons of the
  sequences using one of the phylogenetic methods
• Unfortunately uncertainty is growing in the
  lower levels of the tree, as deletions or
  insertions not easy to recognise
• The challenge is to utilize an appropriate
  combination of sequence weighting, scoring matrix
  and gap penalties, which prevents optimal MSA

5
ClustalW

• This is one of the advanced version of the
  popular and powerful program, where W stand for
  weighting. ClustalW provides more realistic
  alignments that should reflect evolutionary
  changes and more appropriate distribution of gaps
  between conserved domains
• ClustalW performs a global-multiple sequence
  alignment by a different method than MSA,
  although the initial global-multiple sequence
  alignment is calculated similarly
• The steps involved are 1. Pairwise alignment of
  all sequences 2. Use the alignment scores for
  building a phylogenetic tree 3. Progressive
  alignment guided by the phylogenetic
  relationships indicated by the tree
• The most closely related (similar) sequences are
  aligned first, and then additional sequences are
  added
• The initial alignments used to produce the guide
  tree may be obtained by a fast k-tuple approach
  (similar to FASTA) or a slower dynamic
  programming method
• For building a tree genetic distances between
  sequences are calculated as the numbers of
  mismatched positions in an alignment divided by
  the total number of matched positions

6
ClustalW
Sequence A (weight a)
..K Sequence C (weight c)
..L Sequence B (weight b)
..I. Sequence C (weight c)
..L The same procedure applies to other
columns in all pairwise alignments Scores for
matching these two columns in an MSA a x c x
score (K,L ) b x c x score (I,L)
/ncolumnsmpairwisecomparisons
7
An output from ClustalW sequences have
significant similarity
CLUSTAL W (1.82) multiple sequence
alignment gi42542791gbAAH66228.1
MSTAGKVIKCKAAVLWELKKPFSIEEVEVAPPKAHEVRIKMVAAGICRS-
49 gi825623embCAA39813.1
MGTKGKVIKCKAAIAWEAGKPLCIEEVEVAPPKAHEVRIQIIATSLCHT-
49 gi42738724gbAAS42652.1
--MQNFVFRNPTKLIFGKGQ---LEQLKTEIPQFGKKVLLVYGGGSIKRN
45 .
. . .
.   gi42542791gbAAH66228.1
---DEHVVSGNLV-TPLPVILGHEAAGIVESVGEGVTTVKPG--DKVIPL
93 gi825623embCAA39813.1
---DASVIDSKFEGLAFPVIVGHEAAGIVESIGPGVTNVKPG--DKVIPL
94 gi42738724gbAAS42652.1
GIYDNVISILKDINAEVFELTGVEPNPRVSTVKKGIQICKDNGVEFILAV
95 .
. . . . ..
.   gi42542791gbAAH66228.1
FTPQCGKCRICKNPESNYCLKN-DLGNPRG-------------------T
123 gi825623embCAA39813.1
YAPLCRKCKFCLSPLTNLCGKISNLKSPASDQ----------------QL
128 gi42738724gbAAS42652.1
GGGSVIDCTKAIAAGSKYDGDVWDIVTKKAFASEALPFGTVLTLAATGSE
145 . .
. . . . . . . ..
  gi42542791gbAAH66228.1
LQDGTRRFTCSGKPIHHFVGVSTFSQYTVVDENAVAKIDAASPLEKVCLI
173 gi825623embCAA39813.1
MEDKTSRFTCKGKPVYHFFGTSTFSQYTVVSDINLAKIDDDANLERVCLL
178 gi42738724gbAAS42652.1
MNAGSVITNWETNEKYGWGSPVTFPQFSILDPVHTASVPRDQTIYGMVDI
195 . .
. .. . alcohol
dehydrogenase, iron-containing Bacillus cereus
Class I alcohol dehydrogenase, gamma subunit
Homo sapiens Different form of alcohol
dehydrogenase Homo sapiens
8
Localised alignments in sequences

• MSA programs discussed so far are based on
  global alignments, including all available parts
  of sequences
• However many sequences may have blocks of
  similarity, which are separated by low
  similarity regions
• Three approaches were used to develop methods
  more oriented toward this structural feature 1.
  Profile analysis 2. Block analysis 3. Pattern
  searching
• Profiles are found by performing the global MSA
  of a group of sequences and them choosing the
  more highly conserved regions. A score matrix for
  such MSA, called profile, is then made. Once
  produced, the profile is used to search a target
  sequence for possible matches to the profile
  using scores in the table to evaluate the
  likelihood at each position.

9
Profile analysis pattern identification
CONS A B C D.V W Y Z
Gap Len I 8 3 -2
5.21 -18 -6 4 100 100 T
13 19 -5 24.3 -28 14 15
100 100 L 5 5 -5
3.10 -1 5 2 22 22 S
17 14 17 13.1 -8 -15 4
100 100 T 15 3 22
0.9 -22 6 -4 100 100 T
8 -1 12 -219 -15 4 -3
100 100 C 17 0 24
-1.9 -5 14 -7 100 100 V
11 0 18 -1.31 -19 5 -5
100 100 C 10 -8 15
-1115 22 14 -11 100 100 V
7 7 -3 8.26 -24 -6 8
100 100
The profile represents the specific motif pattern
found for the chosen location for a set of hsp70
proteins. It is used to search a target sequence
for matches to the profile. The values are log
odds score of giving the probability of finding
the amino acid in the target sequence at that
position in the profile divided by the
probability of aligning the two aa by random
chance. There are 23 columns, representing 20 aa
1 unknown aa (Z) gap opening and extension
penalties. Gaps are costly unless the profile
itself include gaps, as in the row 3.
10
Profile analysis pattern identification
11
Block Analysis

• This method is very similar to the profile
  search. The major difference is that insertions
  and deletions are not considered. As a result the
  patterns found contain regions of high similarity
  separated by loosely similar or dissimilar
  sequences
• These ungapped patterns may be extracted from
  these aligned regions and used to produce blocks.
  Profile matrices the same as in the previous
  method are built.

Seq1 GVDVLVATPG RLLDLEHQNA ..VKLDQV
EILVLDEADR Seq2 GPDALVSTPG RYLTLEHRNV ..LKPDIV
TIRVLDEADR Seq3 AVEVIVSTPG RLWDLHHQNA
..VQLSQD ELLDLDEADK
Seqn GCDKLNATPG
RLMDLKHQGA ..VKLLFV SILVMDEADR
12
Hidden Markov Models
13
Hidden Markov Model for sequence alignment
14
Hidden Markov Models calculation of transition
probabilities
15
GeneDoc a multiple sequence alignment editor