Transcript No Slide Title

Transfer of information
The main topic of this course is transfer of information.
A month in the lab can easily save you an hour in front
of the computer.
Nothing is impossible for a man who doesn’t have to do
it himself.
But, to err is human, but to really screw things up, you need a computer.
©CMBI 2005
Transfer of information
The main topic of this course is transfer of information.
In the protein world that leads to the questions:
1) From which protein can I transfer information
2) How do I transfer what information from where to wher
Today’s answer is BLAST…
©CMBI 2005
Database Searching with BLAST
Database searching with BLAST involves a series of
topics we will deal with today:
•Database Searching
•Sequence Alignment
•Scoring Matrices
•Significance of an alignment
and:
•BLAST, algorithm
•BLAST, parameters
•BLAST, output
©CMBI 2005
Database Searching
Identify similarities between:
your query sequence
likely with unknown structure and function
database subject sequences
with elucidated structures and function
©CMBI 2005
Database searching concept
The query sequence is compared/aligned with every
subject sequence in the database.
High-scoring database sequences are assumed to be
evolutionary related to the query sequence.
If sequences are related by divergence from a common
ancestor, there are said to be homologous.
We can only transfer information between homologs.
(And we will learn later that that is because structure is maintained longer during evolution than sequence).
©CMBI 2005
Transfer of information
We want to be able to say things like “this serine is
phorphorylated in the database protein, so in my
homologous protein the corresponding serine is likely
to be phosphorylated too”.
That requires that the green serine and the purple serine
both come from a common ancestor that was
phosphorylated too.
And that, in turn, requires that both serines are located
at the same location in their respective structures.
©CMBI 2005
Equivalent structural positions
To know if positions in two different proteins are
equivalent, we need to know both protein structures
and compare them with protein structure comparison
software.
But by the time you have solved one or two protein
structures the four years of your PhD period are over...
So, we need a short-cut, and that, ladies and gentleman,
will be a sequence alignment (i.e. Blast + ...).
©CMBI 2005
Sequence alignment
Sequence alignment is a simple concept. You only have
to find out which pairs of residues in two homologous
sequences are derived from the same residue in the
common ancestor.
TTSASDFRTRTTHIKILLMRL
STSATSYRTRSTHLRLMLMRI seems easy, but:
ASDFTHGTREWDSTYHLIMNV
LTEYSHNSKDFETSFNILLQL looks very hard...
(Still, both alignments seem correct to me, and four weeks from now, you will agree, I hope).
©CMBI 2005
Sequence alignment is easy:
You only need three things:
1) A computer program that produces all possible
alignments, and
2) A computer program that gives each alignment a
score, and, the simplest,
3) A computer program that selects the highest scoring
alignment from the very large number you tried.
(The next two weeks you will learn that only point 2 is difficult)
©CMBI 2005
Scoring Matrix/Substitution Matrix
To score the quality of an alignment you need
‘something’ that compares amino acids, a matrix.
Contains scores for pairs of residues
So, for protein/protein comparisons we need a 20 x 20
matrix of similarity scores where identical amino acids
and those of similar character give higher scores
compared to those of different character.
(And next week you will learn which residues are similar)
©CMBI 2005
Substitution Matrices
Not all amino acids are equal
Residues mutate more easily to similar ones
Residues at surface mutate more easily
Aromatics mutate preferably into aromatics
Mutations tend to favor some substitutions
Core tends to be hydrophobic
Selection tends to favor some substitutions
Cysteines are dangerous at the surface
Cysteines in bridges seldom mutate
©CMBI 2005
PAM250 Matrix
©CMBI 2005
Scoring example
Score of an alignment is the sum of the scores of
all pairs of residues in the alignment
sequence 1: TCCPSIVARSN
sequence 2: SCCPSISARNT
1 12 12 6
2 5 -1 2 6 1 0
=> score = 46
©CMBI 2005
Dayhoff Matrix (1)
The group of Dayhoff created a scoring matrix from a dataset
of closely similar protein sequences that could be aligned
unambiguously.
Then they counted all mutations (and non-mutations) and
calculated the mutation frequencies
With a bit of math, they converted these frequencies into the
famous Dayhoff matrix (also called PAM matrix).
©CMBI 2005
Dayhoff Matrix (2)
Given the frequency of Leu and Val in my sequences, and the frequency of
mutations,, do I see more mutations of V  L than I would expect by chance
alone?
Score of mutation A  B = log (observed a  b mutation
/ expected a  b mutations)
This is called a log odd and can be negative, zero, or positive. Zero means
no information, no contribution to the score of the alignment.
When using a log odds matrix, the total score of the alignment is given by
the sum of the scores for each aligned pair of residues.
©CMBI 2005
Dayhoff Matrix (3)
This log odds matrix is called PAM 1. An evolutionary distance of 1 PAM
(point accepted mutation) means there has been 1 point mutation per 100
residues
PAM 1 may be used to generate matrices for greater evolutionary
distances by multiplying it repeatedly by itself.
PAM250:
– 2,5 mutations per residue.
– equivalent to 20% matches remaining between two sequences,
i.e. 80% of the amino acid positions are observed to have
changed (one or more times).
– is default in many analysis packages.
©CMBI 2005
BLOSUM Matrix
Limit of Dayhoff matrix:
Matrices based on the Dayhoff model of evolutionary rates are
derived from alignments of sequences that are at least 85%
identical; that might not be optimal…
An alternative approach has been developed by Henikoff and
Henikoff using local multiple alignments of more distantly related
sequences.
All matrices are symmetrical...
©CMBI 2005
BLOSUM Matrix (2)
The BLOSUM matrices (BLOcks SUbstitution Matrix) are based on
the BLOCKS database.
The BLOCKS database utilizes the concept of blocks (un-gapped
amino acid pattern), that act as signatures of a family of proteins.
Substitution frequencies for all pairs of amino acids were then
calculated and this used to calculate a log odds BLOSUM matrix.
Different matrices are obtained by varying the identity threshold. For
example, BLOSUM80 was derived using blocks of 80% identity.
©CMBI 2005
Which Matrix to use?
Close relationships (Low PAM, high Blosum)
Distant relationships (High PAM, low Blosum)
BLOSUM 80
PAM 20
BLOSUM 62
PAM 120
More conserved
Often used defaults are: PAM250, BLOSUM62
BLOSUM 45
PAM 250
More variable
Significance of alignment (1)
When is an alignment statistically significant?
In other words:
How much different is the alignment score found from scores
obtained by aligning any odd sequences to the query sequence?
Or:
What is the probability that an alignment with this score could have
arisen by chance?
©CMBI 2005
Significance of alignment (2)
Database size= 20 x 106 amino acids
peptide
#hits
A
AP
IAP
LIAP
WLIAP
KWLIAP
KWLIAPY
1 x 106
50000
2500
125
6
0,3
0,015
©CMBI 2005
BLAST
Question: What database sequences are most similar to
(or contain the most similar regions to) my own sequence?
•BLAST finds the highest scoring locally optimal
alignments between a query sequence and all database
sequences.
•Very fast algorithm
•Can be used to search extremely large databases
•Sufficiently sensitive and selective for most purposes
•Robust – the default parameters can usually be used
©CMBI 2005
BLAST – Algorithme
Step 1: Read/understand user query sequence.
Step 2: Use hashing technology to select several thousand
likely candidates.
Step 3: Do a real alignment between the query sequence
and those likely candidate. ‘Real alignment’ is a main topic
of this course.
Step 4: Present output to user.
©CMBI 2005
BLAST Algorithm, Step 2
The program first looks for series of short, highly similar
fragment, it extends these matching segments in both
directions by adding residues. Residues will be added
until the incremental score drops below a threshold.
©CMBI 2005
Basic BLAST Algorithms
Program
Query
Database
BLASTP
Protein
Protein
BLASTN
DNA
DNA
BLASTX
translatedDNA
protein
TBLASTN
protein
translatedDNA
TBLASTX
translatedDNA
translatedDNA
©CMBI 2005
PSI-BLAST
Position-Specific Iterated BLAST
• Distant relationships are often best detected by motif
or profile searches rather than pair-wise comparisons
• PSI-BLAST first performs a BLAST search.
• PSI-BLAST uses the information from significant
BLAST alignments returned to construct a position
specific score matrix, which replaces the query
sequence for the next round of database searching.
• PSI-BLAST may be iterated until no new significant
alignments are found.
©CMBI 2005
BLAST Input
Steps in running BLAST:
•Entering your query sequence (cut-and-paste)
•Select the database(s) you want to search
And, optionally:
•Choose output parameters
•Choose alignment parameters (scoring matrix, filters,….)
Example query=
>something
AFIWLLSCYALLGTTFGCGVNAIHPVLTGLSKIVNGEEAVPGTWPWQVTLQDRSGFHFC
GGSLISEDWVVTAAHCGVRTSEILIAGEFDQGSDEDNIQVLRIAKVFKQPKYSILTVNND
ITLLKLASPARYSQTISAVCLPSVDDDAGSLCATTGWGRTKYNANKSPDKLERAALPLLT
NAECKRSWGRRLTDVMICGAASGVSSCMGDSGGPLVCQKDGAYTLVAIVSWASDTCSASS
GGVYAKVTKIIPWVQKILSSN
©CMBI 2005
BLAST Output
A high score
indicates a likely
relationship
A low probability
indicates that a
match is unlikely to
have arisen by
chance
©CMBI 2010
BLAST Output
Low scores with high
probabilities suggest
that matches have
arisen by chance
©CMBI 2010
Alignment Significance in BLAST
P-value (probability)
Relates the score for an alignment to the likelihood that it
arose by chance. The closer to zero, the greater the
confidence that the hit is real.
E-value (expect value)
The number of alignments with E that would be expected
by chance in that database (e.g. if E=10, 10 matches with
scores this high are expected to be found by chance).
A match will be reported if its E is below the threshold.
Lower E thresholds are more stringent, and report fewer
matches.
©CMBI 2005
BLAST result: easy
©CMBI 2010
BLAST result: less easy
©CMBI 2010
BLAST result: very difficult
©CMBI 2010
Low complexity filter
Many sequences contain repeats or stretches that consist
predominantly of one type of amino acid.
E.g. Many nuclear proteins have a poly-asparagine tail,
membrane proteins often consist of mainly hydrophobic
amino acids, or many binding proteins have proline rich
stretches.
ASDFGTRGHPPPPPPPPPPP--------------NPPPPPPPPPLTSSDFRGT
Are NOT homologs, but analogs.
©CMBI 2005
BLAST - Low complexity filter
NNNNNNNN
Your BLAST query sequence will look like this:
NNNNNNNN
Filter ON
Filter OFF
©CMBI 2010
Demo
IJs, CNCZ, en het internet dienende komt nu een demo…
©CMBI 2005