Searching Sequence
Databases
© Wiley Publishing. 2007. All Rights Reserved.
Learning Objectives
Finding out why similarity searches are so
important
Understanding the relationship between
homology, similarity, and identity
Being able to run a BLAST and to interpret
program output
Understanding the concept of e-values
Knowing how to ask biological questions with
BLAST
Outline
Biological meaning of sequence
similarity
Homology, identity, and similarity
Running BLAST
Interpreting a BLAST output
Making a biological analysis with
BLAST
Running PSI-BLAST the latest BLAST
version
Sequence Similarity
Two protein sequences with more than 25 % identity (over
100 amino acids ) are homologues
Two DNA sequences with more than 70 % identity (over
100 nucleotides) are homologues
Homologous sequences have
• A common ancestor (proteins and DNA)
• A similar 3D structure (proteins)
• Often a similar function (proteins)
Homology
When two proteins have less than 25% identity
• They can be homologous or non-homologous
• Within this range of identity, it’s impossible to say which
is true
This range of identity is called the “Twilight Zone”
Homology, Similarity,
and Identity
Identity is a measure made on an alignment
• Sequence A can be “32 % identical to” Sequence B
Similarity is a measure of how close two amino acids are to
identical
• For instance, isoleucine and leucine are similar
Homology is a property that exists or does not exist
• Sequence A IS or IS NOT homologous to Sequence B
• Sequence A cannot be “40% homologous to” B
Homology is established on the basis of measured similarity
or identity
How to Establish Homology
Compare Protein A with every other protein in a database such as
Swiss-Prot
Identify a Protein B that is 40% identical to your protein
• Specialists prefer using E-values but the idea is the same (more on this in a
minute)
You can conclude that A and B are probably homologous if they are very
similar
• It’s like saying, “John and Nancy are probably brother and sister because
they are very similar.”
If you know the structure or the function of B, then A and B probably
have the same structure
In-silico Biology
When establishing that two
proteins (A and B) are
homologous, you can
extrapolate everything you
know from one to the
other.
It’s like making a virtual
experiment.
This is in-silico biology!
BLAST
BLAST: Basic Local Alignment Search Tool
BLAST is a tool for comparing one sequence with
all the other sequences in a database
BLAST can compare
• DNA sequences
• Protein sequences
BLAST is more accurate for comparing protein
sequences than for comparing DNA sequences
BLAST (cont’d.)
BLAST makes local alignments
• It only aligns what can be aligned
• It ignores the rest
BLAST is very fast
• You need only a few minutes to search Swiss-Prot
on a standard PC
Many BLAST flavors are available for a variety
of tasks
Many BLAST Flavors . . .
BLASTing a Protein Sequence
Running blastp
Choose one of the public servers
• NCBI [Link]/blast
• EBI [Link]/blast
• EMBNet [Link]/blast
Select a database to search:
• NR to find any protein sequence
• Swiss-Prot to find proteins with known functions
• PDB to find proteins with known structures
Cut and paste your sequence
Click the BLAST button
Reading BLAST Output
Graphic Display
• Overview of the alignments
Hit List
• Gives the score of each
match
Alignments
• Details of each alignment
The Graphic Display
The Horizontal Axis (0-700)
corresponds to your protein
(query)
Color codes indicate that
match’s quality
• Red: very good
• Green: acceptable
• Black: bad
Thin lines join independent
matches on the same
sequence
The Hit List
Sequence accession
number
• Depends on the database
Description
• Taken from the database
Bit score
• High bit score = good match
E-Value
• Low E-value = good match
Links
• Genome
• Uniref, database of
transcripts
The E-Values
E-value means expectation value
The E-value is the measure most commonly used for estimating
sequence similarity
How many times is a match at least as good expected to happen by
chance ?
• This estimate is based on the similarity measure
If a match is highly unexpected, it probably results from something
other than chance
• Common origin is the most likely explanation
• This is how homology is inferred
Which Value for Your
E-Values ?
Low E-value good hit
• 1 = bad e-Value
• 10e-3 = borderline E-value
• 10e-4 = good E-value
• 10e-10 = very good E-value
E-values lower than 10e-4 indicate possible homology
E-values higher than 10e-4 require extra evidence to
support homology
Why Use E-Values?
E-values make it possible to compare alignment of
different lengths
E-values are used by most sequence comparison programs
• PSI-BLAST
• Domain Search
• FASTA
E-values always have the same meaning
• You can compare the output of different programs
The Alignments
Look for clusters of
identity
Gray residues are low-
complexity regions
Grayed-out regions
have been removed
from your sequence to
avoid false hits
BLASTing DNA Sequences
The BLAST program you need depends on your DNA
sequence
• Coding DNA
• Non Coding DNA
BLASTing DNA sequences is less accurate than BLASTing
protein sequences
If your sequence is coding, blastx and tblastx will
translate it for you on its 6 possible reading frames
BLASTing DNA Sequences
Asking the Right Question
with BLAST
The BLAST Way of
Doing Things
The original BLAST paper is the fourth-most-cited
scientific publication
• 21,000 citations for BLAST
• 18,000 citations for PSI-BLAST
BLAST has changed many aspects of modern
biology
The following slides show more BLAST
procedures
• They are not necessarily the best procedures
• They are effective ways of getting the job done on the
spot
Gene-Hunting with BLAST
Predicting a Protein Function Cut your genome sequence in little
(2~5Kb) overlapping sequences. Use
blastx to BLAST each piece of genome
against NR (the Non Redundant protein
database). This works better if you
have no introns (bacteria).
The complicated alternative is to run
gene-prediction software program.
In-silico Analysis with BLAST
Use blastp to BLAST your protein
sequence against SWISS-PROT. If you
Predicting a Protein Function get a good hit (more than 25 percent
identity) over the complete length of
the protein, you’ve solved your problem
and you know that your protein has the
same function as the SWISS-PROT
protein.
The complicated alternative is to
conduct domain analysis or wet-lab
experiments
Structural Analysis with BLAST
Predicting a Protein 3D StructureUse blastp to BLAST your protein
against PDB (the database of protein
structure). If you get a good hit (more
than 25 percent identity), you know that
your protein and this good hit have a
similar 3-D structure.
The complicated alternative is to do
Homology Modeling, X-ray or NMR
analysis of your protein
Gathering Members of a
Protein Family
Finding Protein Family Members Use blastp (or its more powerful cousin
PSI-BLAST) and run it against NR (the
non-redundant protein family). After
you have all the members of the family,
you can make a multiple-sequence
alignment (see Chapter 9) and draw a
phylogenetic tree.
The complicated alternative is to use
PCR for cloning your sequences
Some Reasons for Changing
the Default Parameters
PSI-BLAST
PSI-BLAST is Position-Specific Iterated BLAST
• More sensitive than BLAST: finds matches BLAST would not find
• More specific than BLAST: reports fewer false matches
• A bit slower than BLAST
PSI-BLAST finds remote homologues
• Will let you identify very distant members of your protein family
PSI-BLAST uses the results of each iteration to increase
its specificity
PSI-BLAST Iterations
PSI-BLAST uses the best
results of the first iteration
to build a profile (PSSM)
PSI-BLAST uses the profile
to re-scan the database
PSI-BLAST keeps re-
scanning until it stops
finding new matches
Some Tips for Using PSI-BLAST
If your protein is multi-domain, search one
domain at a time
PSI-BLAST is slower than normal BLAST
because of the iterations
You can feed PSI-BLAST with your own
PSSM
• Use the NCBI server for this purpose
Going Farther
Each BLAST online server is unique
Shop around to find the right database
If you need to look for exact matches between a sequence and a
genome use BLAT
• No it’s not a typo
• You can find it at [Link]
If you want something more accurate than BLAST, use Smith and
Waterman
• It’s also slower than BLAST
• You can find it at [Link]