TGAC Pairwise Sequence Comparisons

1
TGACPairwise SequenceComparisons
The Artist in His Museum, Charles Wilson Peale
2
Why Compare Sequences?

• Determine whether two sequences are homologous
• Determine how two sequences are related
• Find common evolutionary histories
• Find common functional domains

3

• Identity
• Similarity
• Homology
• Paralogy
• Orthology

4
Types of Homology
5
Comparing Sequences

• Dotplots
• Alignments
• Scoring alignments
• Algorithms for finding alignments
• Dynamic programming
• Heuristic methods
• BLAST
• FASTA

6
Comparing Sequences

• Dotplots
• Alignments
• Scoring alignments
• Algorithms for finding alignments
• Dynamic programming
• Heuristic methods
• BLAST
• FASTA

7
S
T
O
P
S
S
T
O
O
P
8
S
T
O
P
S
S
T
O
P
S
9
S
T
O
P
S
S
P
O
T
S
10
a
c
t
g
a
t
c
g
a
a
c
t
g
a
c
t
11
a
c
t
g
a
t
c
g
a
g
t
c
a
g
t
c
12
a
c
t
g
a
t
c
g
a
a
c
t
g
g
t
c
13
t
g
t
c
t
g
a
g
a
a
a
c
t
g
t
c
t
a
g
t
g
a
a
14
gt
tc
ct
ta
ga
ag
gt
tg
aa
aa
ac
c
ta
gt
tc
ct
ta
ag
gt
tg
ga
aa
a
15
Self-Alignment of Human LDL Receptor
16
Dot Plot Analysis

• Plot two sequences or sequence against itself
• Diagonal is identical match
• Variable word size possible
• Good for finding features
• Duplications
• Inversions
• Deletions

17
Dot Plot Programs

• EMBOSS has several programs (available via PISE)
• Dotter
• Dotlet (web-based)

18
Comparing Sequences

• Dotplots
• Alignments
• Scoring alignments
• Algorithms for finding alignments
• Dynamic programming
• Heuristic methods
• BLAST
• FASTA

19
Hamming Distance
Between strings of equal length, the number of
mismatches
fatcat fatrat
Hamming Distance 2
fatcats fastcat
Hamming Distance 5
20
Alignment
fatcats fastcat
Hamming Distance 5
fa-tcats fastcat-
Edit Distance 2
21
Optimal Alignment

• Given two sequences, there are many possible
  alignments.
• The edit distance is the minimal number of
  differences in an alignment.
• We can also use a more complicated scoring system
  to evaluate alignments.

ac-gtac-g
acg-tac-g
acagtatag
acagtatag
acgtac----g
-ac-agtatag
22
Comparing Sequences

• Dotplots
• Alignments
• Scoring alignments
• Algorithms for finding alignments
• Dynamic programming
• Heuristic methods
• BLAST
• FASTA

23
Scoring Alignments

• Additive
• Score of AB Score of A Score of B
• Biologically meaningful high score
• Treat gaps differently than mismatches
• Point mutations are more common than insertions
  or deletions.

24
Scoring Systems

• Primitive option Aligned residues are either
  identical or not identical
• Physically motivated Match score is
  proportionate to chemical similarity
• Mutationally motivated (PAM, BLOSUM)If
  mutations often convert one residue to another,
  give that a higher match score.

25
PAM(Point/Percent Accepted Mutation)

• One PAM unit is an average of 1 mutation per 100
  amino acids
• PAM 1 was generated from 71 protein sequence
  groups of at least 85 similarity to avoid errors
  from multiple mutations at the same site, as well
  as insertions and deletions
• Convert mutation probabilities to a (log odds)
  scoring matrix

26
PAM Matrices

• PAMn supplies scoring for sequences that have
  diverged n PAM units
• PAM1n generates other matrices based on Markov
  model
• PAM250 represents an average of 250 mutations in
  100 amino acids

27
BLOSUM(BLOck SUbstitution Matrix)

• BLOCKS is a database of 3,000 blocks short,
  continuous multiple alignments of protein domains
  (functional sequences)
• Generates substitution frequency, which can be
  converted into log odds score
• BLOSUM62
• Block has 62 similarity
• Close to PAM250
• More Reliable

28
Gap Penalties

• Insertions and deletions are less likely than
  substitutions gaps should be penalized.
• Simple gap penalty takes a penalty for each match
  of gap to residue.
• Lots of small deletions are less biologically
  likely than one long deletion.
• Affine gap penalty has different penalties for
  opening a gap or for continuing a gap.
• Opening a gap (5 nucleotides, 11 proteins)
• Continuing a gap (2 nucleotides, 1 proteins)

29
Finding best alignment

• We know how to recognize optimal alignment in a
  pool of possibilities
• We could enumerate all possibilities, then choose
  the best one.
• Thats slow. Is it necessary?

30
Comparing Sequences

• Dotplots
• Alignments
• Scoring alignments
• Algorithms for finding alignments
• Dynamic programming
• Heuristic methods
• BLAST
• FASTA

31
Premise of Dynamic Programming

• If we know the optimal alignment of the prefixes
  of two sequences, we can find the optimal
  alignment of the entire sequence.

32
Dynamic Programming Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
-
G
GT
GTA
33
Initializing the Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
GAT --- -9
G - -3
GA -- -6
empty alignment 0
-
- G -3
G
-- GT -6
GT
--- GTA -9
GTA
34
Filling in the Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
G - -3
empty alignment 0
-
- G -3
G G -3
G
-- GT -6
GT
--- GTA -9
GTA
35
Filling in the Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
G - -3
empty alignment 0
-
- G -3
G G -3
G
-- GT -6
GT
--- GTA -9
GTA
36
Filling in the Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
GAT --- -9
GA -- -6
G - -3
G - -3
empty alignment 0
-
- G -3
G G -3
GA G- 0
GAT G-- -3
G
-- GT -6
G- GT 0
GAT G-T 2
GA GT 2
GT
--- GTA -9
G-- GTA 0
GAT GTA 1
G-A GTA 3
GTA
37
Filling in the Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
-3
-6
-9
0
-
-3
-3
0
-3
G
-6
0
2
2
GT
0
-9
3
1
GTA
38
Filling in the Matrix
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
G
GA
GAT
-
-3
-6
-9
0
-
-3
-3
0
-3
G
-6
0
2
2
GT
0
-9
3
1
GTA
39
Finding the Alignment
Match 3 Mismatch -1 Gap -3
Sequence A GAT Sequence B GTA
GAT GTA
G
GA
GAT
-
-3
-6
-9
0
-
-3
-3
0
-3
G
-6
0
2
2
GT
0
-9
3
1
GTA
40
A larger example
http//meetings.cshl.org/tgac/tgac/flash/justMatri
x.swf
41
Local Alignments
a_fat_cat_lives_a_happy_existence the_dog_likes_to
_chase_cats
A local alignment is an alignment between a
substring of one sequence and a substring of the
other.
42
Dynamic Programming

• Global alignment (what we just did)Initiate
  traceback in bottom right-hand corner of the
  matrix.Often called Needleman-Wunsch
• Local alignment Find top-scoring
  sub-alignments.Optimal scores cannot be
  negative. Initiate traceback at highest scoring
  cell in the matrix.Often called Smith-Waterman

43
Dynamic Programming

• Provides an optimal alignment
• Results depend on scoring system and gap
  penalties
• Sometimes too slow
• Perfectly good for aligning two sequences
• Slow for aligning query sequence against whole
  database of sequences

44
Pairwise Sequence Alignment

• Often, we dont know which two sequences to
  align.
• Given a query sequence, we need a way to find all
  homologous sequences in a large database.

45
Comparing Sequences

• Dotplots
• Alignments
• Scoring alignments
• Algorithms for generating alignments
• Dynamic programming
• Heuristic methods
• BLAST
• FASTA

46
BLAST

• Basic Local Alignment Search ToolFinds local
  alignments.
• Most popular tool available to search a database
  for sequences homologous to a query sequence.
• Will not necessarily find the optimal alignment,
  if it has a low score.
• Will find all high-scoring alignments with high
  probability

47
BLAST Steps

• -1. Preprocess database to get sorted list of
  words in every sequence in the database.
• For each word of length W in the query, generate
  a list of all possible words with a score of at
  least threshold T
• Generate hits by searching the database for any
  words on the list.
• Use dynamic programming to extend hits until the
  score drops gtX return highest-scoring segment
  pair (HSP)
• Evaluate significance of extended hits

48
Query word (W 3)
Query GSDFWQETRASFGCSLAALLNKCKTPQGQRLVNQWIKQPLMDK
NRIEERLNLVEAFGCATSWPI
PQG 18 PEG 15 PRG 14 PKG 14 PNG 13 PDG 13 PHG 13 P
MG 13 PSG 13 PQA 12 PQN 12
Neighborhood words
Neighborhood score threshold (T 13)
Hit
Query SLAALLNKCKTPQGQRLVNQWIKQPLMDKNRIEERLNLVEA
LAL TP G R W P D ER
A Subject TLASVLDCTVTPMGSRMLKRWLHMPVRDTRVLLERQQT
IGA
High-scoring Segment Pair (HSP)
49
BLAST Settings

• Program
• Filters
• Expect
• Word Size
• Show
• Number of Descriptions/Alignments

50
BLAST Programs
51
Other BLAST Programs

• MEGABLAST
• Genomic BLAST
• Fugu rubripes

52
BLAST Databases
53
BLASTing Custom Databases

• If you run BLAST on your own computer, instead of
  at NCBI, you can create a custom database.
• Gather sequences for database (e.g., by using
  Batch ENTREZ)
• Run formatdb to preprocess the database.

54
BLAST Filters

• Low complexity
• Human repeats
• Mask for lookup table only
• Mask lower case
• Limit by Entrez query

55
ENTREZ Queries

• nucleotide allFilter NOT specimen-voucherAll
  Fields
• You can eliminate most of the BAC-type records
  from the default nucleotide database with the
  following query nucleotide allFilter NOT
  htgKeyword

56
E-values

• Lower values give more stringent results
• Default is 10 fractional values permissible
• E10 means 10 matches are expected by chance
• In output, E lt 0.05 is significant

57
BLAST Output

• Reference
• Query
• Database
• TaxBLAST link
• Graphical Overview aligned matches

58
BLAST Output Significant Alignments

• SeqID Genbank ID and Accession
• Description (links to entry)
• Score (links down)
• E-value
• LocusLink
• Unigene

59
BLAST Output - Alignments

• Length
• Score
• Raw score is just score from scoring matrix
• Bit score is adjusted for statistical properties
  of scoring matrix, and can be compared between
  searches.
• E-value
• Identities/Gaps
• Strand (e.g. Plus/Plus)
• Alignment shows identity

60
BLAST Output - Proteins

• Conserved Domains Pfam CD-Search
• Alignments
• Matching letters shown
• Mismatches blank
• Conservative substitutions shown by
• Gaps are shown by dashes

61
BLAST Output End Summary

• Database -- of letters sequences
• Lambda, K, H
• Gapped Lambda, K, H
• Matrix
• Gap Penalties
• Hit summary , extensions, gt 10.0
• HSPs
• length of query
• T,A, X1, X2, X3, S1, S2

62
Gapped BLAST

• Find two non-overlapping hits of score at least T
  and distance at most d one from another
• Ungapped extension
• If the HSP generated has high enough normalized
  score, start gapped extension
• Report resulting alignment if it has sufficiently
  large E-value

63
Benefits of Protein Searches

• 20 amino acids reduce chance of random matches
• Protein DBs are smaller
• Degrees of similarity
• Chemical
• Observed mutation frequencies
• Mutational steps between codons

64
FASTA

• Find hot-spots (pairs of words of length k these
  are called hits in BLAST)
• Locate sequences of consecutive hot spots on a
  diagonal
• Combine sub-alignments into a longer alignment
• Find alternative local alignments using dynamic
  programming restricted to a ribbon along the
  diagonal containing best run

65
FASTA Algorithm
66
FASTA Algorithm
67
FASTA Programs

• Fasta3 - Compare a protein sequence to a protein
  database, or a DNA sequence to a DNA database
• Ssearch3 - Compare a protein sequence to a
  protein database, or a DNA database, using the
  Smith-Waterman algorithm. It is very slow but
  much more sensitive for full-length protein
  comparison.
• Fastx3 - Compare a DNA sequence to a protein
  database, by comparing the translated DNA
  sequence in three frames and allowing gaps and
  frameshifts.

68
BLAST and FASTA

• BLAST for proteins and speed
• FASTA for DNA and frameshifts
• FASTA for accurate statistics (protein and coding
  DNA)
• SSEARCH for optimal

69
Blast vs. Fasta II

• Database format is the same
• BLAST is generally faster
• FASTA may produce better alignments
• blastp will show several alignments between
  domains in the same sequences. Fasta3 shows one
  alignment for each sequence pair.

70
Blast vs. Fasta III

• Different default matrices and gap penalties
• BLAST cannot search very short sequences (can use
  Fasta, but EMBOSS fuzznuc or fuzzpro are better)
• Fasta searches one strand of DNA
• Fasta does not filter low complexity by default

71
Sites using WU-BLAST

• European Bioinformatics Institute BLAST server
• EMBL Advanced BLAST2 Search
• Institut Pasteur
• Berkeley Drosophila Genome Project (BDGP)
• European Drosophila Genome Project at the EBI
• Mendel Bioinformatics Group at the John Innes
  Centre
• Mouse Genome Database at the Jackson Laboratory
• PlasmoDB (Plasmodium Genome Resource) at the
  University of Pennsylvania
• Saccharomyces cerevisiae search at Stanford
  University
• TAP (Transcript Assembly Program) by Zhengyan
  ("George") Kan
• TIGR Gene Indices
• TIGR Microbial Genomes Database
• WU Genome Sequencing Center