Transcript Protein Domains/Motifs

Previous Lecture: Multiple Alignment
This Lecture
Introduction to Biostatistics and Bioinformatics
Motifs
Learning Objectives
•
•
•
•
•
Restriction sites
Finding genes in DNA sequences
Regulatory sites in DNA
Protein signals (transport and processing)
Protein functional domains & motif
databases
• Regular Expressions
• Position Specific Scoring Matrix
& Hidden Markov Models
Restriction Sites
• Bacteria make restriction
enzymes that cut DNA at
specific sequences
(4-8 base patterns)
• Very simple to find these patterns - can even
use the “Find” function of your web browser
or word processor
• Open any page of text and look for “CAT”
– you now have a restriction site search program!
NEBcutter2
http://tools.neb.com/NEBcutter2/
Finding Genes in
Genomic DNA
• Translate (in all 6 reading frames) and look
for similarity to known protein sequences
• Look for long Open Reading Frames (ORFs)
between start and stop codons
(start=ATG, stop=TAA, TAG, TGA)
• Look for known gene markers
•
TAATAA box, intron splice sites, etc.
• Statistical methods (codon preference)
GCCACATGTAGATAATTGAAACTGGATCCTCATCCCTCGCCTTGTACAAAAATCAACTCCAGATGGATCTAAG
ATTTAAATCTAACACCTGAAACCATAAAAATTCTAGGAGATAACACTGGCAAAGCTATTCTAGACATTGGCTT
AGGCAAAGAGTTCGTGACCAAGAACCCAAAAGCAAATGCAACAAAAACAAAAATAAATAGGTGGGACCTGATT
AAACTGAAAAGCCTCTGCACAGCAAAAGAAATAATCAGCAGAGTAAACAGACAACCCACAGAATGAGAGAAAA
TATTTGCAAACCATGCATCTGATGACAAAGGACTAATATCCAGAATCTACAAGGAACTCAAACAAATCAGCAA
GAAAAAAATAACCCCATCAAAAAGTGGGCAAAGGAATGAATAGACAATTCTCAAAATATACAAATGGCCAATA
AACATACGAAAAACTGTTCAACATCACTAATTATCAGGGAAATGCAAATTAAAACCACAATGAGATGCCACCT
TACTCCTGCAAGAATGGCCATAATAAAAAAAAATCAAAAAAGAATAAATGTTGGTGTGAATGTGGTGAAAAGA
GAACACTTTGACACTGCTGGTGGGAATGGAAACTAGTACAACCACTGTGGAAAACAGTACCGAGATTTCTTAA
AGAACTACAAGTAGAACTACCATTTGATCCAGCAATCCCACTACTGGGTATCTACCCAGAGGAAAAGAAGTCA
TTATTTGAAAAAGACACTTGTACATACATGTTTATAGCAGCACAATTTGCAATTGCAAAGATATGGAACCAGT
CTAAATGCCCATCAACCAACAAATGGATAAAGAAAATATGGTATATATACACCATGGAACACTACTCAGCCAT
AAAAAGGAACAAAATAATGGCAACTCACAGATGGAGTTGGAGACCACTATTCTAAGTGAAATAACTCAGGAAT
GGAAAACCAAATATTGTATGTTCTCACTTATAAGTGGGAGCTAAGCTATGAGGACAAAAGGCATAAGAATTAT
ACTATGGACTTTGGGGACTCGGGGGAAAGGGTGGGAGGGGGATGAGGGACAAAAGACTACACATTGGGTGCAG
TGTACACTGCTGAGGTGATGGGTGCACCAAAATCTCAGAAATTACCACTAAAGAACTTATCCATGTAACTAAA
AACCACCTCTACCCAAATAATTTTGAAATAAAAAATAAAAATATTTTAAAAAGAACTCTTTAAAATAAATAAT
GAAAAGCACCAACAGACTTATGAACAGGCAATAGAAAAAATGAGAAATAGAAAGGAATACAAATAAAAGTACA
GAAAAAAAATATGGCAAGTTATTCAACCAAACTGGTAATTTGAAATCCAGATTGAAATAATGCAAAAAAAAGG
CAATTTCTGGCACCATGGCAGACCAGGTACCTGGATGATCTGTTGCTGAAAACAACTGAAAATGCTGGTTAAA
ATATATTAACACATTCTTGAATACAGTCATGGCCAAAGGAAGTCACATGACTAAGCCCACAGTCAAGGAGTGA
GAAAGTATTCTCTACCTACCATGAGGCCAGGGCAAGGGTGTGCACTTTTTTTTTTCTTCTGTTCATTGAATAC
AGTCACTGTGTATTTTACATACTTTCATTTAGTCTTATGACAATCCTATGAAACAAGTACTTTTAAAAAAATT
GAGATAACAGTTGCATACCGTGAAATTCATCCATTTAAAGTGAGCAATTCACAGGTGCAGCTAGCTCAGTCAG
CAGAGCATAAGACTCTTAAAGTGAACAATTCAGTGCTTTTTAGTATATTCACAGAGTTGTGCAACCATCACCA
CTATCTAATTGGTCTTAGTCTGTTTGGGCTGCCATAACAAAATACCACAAACTGGATAGCTCATAAACAACAG
GCATTTATTGCTCACAGTTCTAGAGGCTGGAAGTGCAAGATTAAGATGCCAGCAGATTCTGTGTCTGCTGAGG
GCCTGTTCCTCATAGAAGGTGCCCTCTTGCTGAATTCTCACATGGTGGAAGGGGGAAAACAAGCTTGCATTGC
Intron/Exon structure
• Gene finding programs work well in bacteria
• None of the gene prediction programs do a
very good job of predicting eukaryotic
intron/exon boundaries
• The only reasonable gene models are based
on alignment of cDNAs to genome sequence
• >50% of all human genes still do not have an
accurate coding sequence defined
(transcription start, intron splice sites)
Gene Finding on the Web
GRAIL: Oak Ridge Natl. Lab, Oak Ridge, TN
–
http://compbio.ornl.gov/grailexp
ORFfinder: NCBI
– http://www.ncbi.nlm.nih.gov/gorf/gorf.html
DNA translation: Univ. of Minnesota Med. School
– http://alces.med.umn.edu/webtrans.html
GenLang
– http://cbil.humgen.upenn.edu/~sdong/genlang.html
BCM GeneFinder: Baylor College of Medicine, Houston, TX
– http://dot.imgen.bcm.tmc.edu:9331/seq-search/gene-search.html
– http://dot.imgen.bcm.tmc.edu:9331/gene-finder/gf.html
Truth?
• There may not be a "correct" answer to the gene
finding problem
• Some genes have more than one start and stop
position on the DNA
• Alternative splicing
(a portion of the DNA is sometimes in an exon, sometimes in an
intron)
• Pseudogenes - look like genes, but no longer
function
• All computational gene predictions need to be
experimentally verified (RNA-seq!!)
Genomic Sequence
• Once each gene is located on the
chromosome, it becomes possible to get
upstream genomic sequence
• This is where transcription factor (TF)
binding sites are located
– promoters and enhancers
• Search for known TF sites, and discover
new ones (among co-regulated genes)
Phage CRO repressor bound to DNA
Andrew Coulson & Roger Sayles with RasMol, Univ. of Edinburgh 1993
Sequence
Logos
Many DNA Regulatory Sequences
are Known
– JASPAR: a curated, non-redundant set of
transcription factor binding sites from published articles
(currently 593 non-redundant matrics).
– UniProbe: binding sites of transcription factors
determined by in vitro protein binding microarray
(data for 406 DNA binding proteins on all k-mers)
– TransFac
• Became a private for profit company (BIOBASE/Quiagen)
• Stopped adding new entries to public data in 2005
– The Eukaryotic Promoter Database (EPD)
• 1314 entries taken directly from scientific literature
JASPAR page for CTCF
Position Scoring Matrix
Biopython Bio.motifs package (similar to BioPerl TFBS)
Count matrix:
A:
C:
G:
T:
0
4.00
16.00
0.00
0.00
1
19.00
0.00
1.00
0.00
2
0.00
20.00
0.00
0.00
3
0.00
0.00
20.00
0.00
4
0.00
0.00
0.00
20.00
5
0.00
0.00
20.00
0.00
Normalized position weight matrix
(with pseudocounts) = probability of each base
A:
C:
G:
T:
0
0.22
0.59
0.09
0.09
1
0.69
0.09
0.12
0.09
2
0.09
0.72
0.09
0.09
3
0.09
0.09
0.72
0.09
4
0.09
0.09
0.09
0.72
5
0.09
0.09
0.72
0.09
Position Specific Scoring Matrix
(log odds ratios of matrix vs background):
0
A:
C:
G:
T:
1
-0.19
1.25
-1.42
-1.42
2
1.46
-1.42
-1.00
-1.42
3
-1.42
1.52
-1.42
-1.42
4
-1.42
-1.42
1.52
-1.42
5
-1.42
-1.42
-1.42
1.52
-1.42
-1.42
1.52
-1.42
Positive scores show that a base is more likely to come from the motif,
negative scores are more likely to come from background
>>> m.consensus
Seq('CACGTG', IUPACUnambiguousDNA())
>>>m.weblogo("mymotif.png")
Motif Search Methods
Exact Match
>>> match = seq.count('CACGTG')
Regular Expression Match
>>> match = re.search(r'[CA][AG]CG[TC]G', seq)
PSSM Search
>>> from Bio import motifs
>>> for position, score in pssm.search(seq, threshold=7.0):
...
print("Position %d: score = %5.3f" % (position, score))
...
Threshold of log-odds 7 = 100x more likely to
Position 0: score = 5.622
Position -20: score = 4.601
occur in motif than random background
Position 10: score = 3.037
Negative positions are on - strand
Position 13: score = 5.738
A highly selective motif should only match once
(or zero times) in each sequence tested.
DE
SQ
SF
ST
BF
IFI-6-16 (interferon-induced gene 6-16); G000176.
gGGAAAaTGAAACT
-127
-89
T00428 ISGF-3; Quality: 6; Species: human, Homo sapiens.
TF Binding sites lack information
• Most TF binding sites are determined by just a few base
pairs (typically 6-12)
• Sequence is variable (consensus)
• This is not enough information for proteins to locate
unique promoters for each gene in a 3 billion base
genome
• TF's bind cooperatively and combinatorially
– The key is in the location in relation to each other and to the
transcription units of genes + epigenetic factors
• Can use phylogenetic conservation to help predict binding sites
Web tools for TFBS
Promoter Scan: NIH Bioinformatics (BIMAS)
http://www-bimas.cit.nih.gov/molbio/proscan/
Signal Scan: NIH Bioinformatics (BIMAS) – uses old TransFac database
http://www-bimas.cit.nih.gov/molbio/signal/
TFSEARCH (uses 1998 version of TransFac)
http://www.cbrc.jp/research/db/TFSEARCH.html
JASPAR (search motifs in one sequence), ConSite
http://jaspar.genereg.net/
http://consite.genereg.net/
Toucan workbench for regulatory sequence analysis
https://gbiomed.kuleuven.be/english/research/50000622/lcb/tools/toucan
TargetFinder: Telethon Inst.of Genetics and Medicine, Milan, Italy
http://www.targetfinder.org/index.php/findtargets
RSAT: Regulatory Sequence Analysis Toolkit
http://rsat.ulb.ac.be/rsat/
MotifMogul: A web server that enables the analysis of multiple DNA sequences with PWM from
JASPAR and TRANSFAC using 3 different algorithms (CLOVER, MotifLocator, MotifScanner)
http://xerad.systemsbiology.net/MotifMogulServer/index.html
Protein Sequence
Protein Sequence Analysis
• Molecular properties (pH, mol. wt.
isoelectric point, hydrophobicity)
• Motifs (signal peptide, coiled-coil, transmembrane, etc.)
• Protein Families
• Secondary Structure (helix vs. beta-sheet)
• 3-D prediction, Threading
Chemical Properties of
Proteins
• Proteins are linear polymers of 20 amino
acids
• Chemical properties of the protein are
determined by its amino acids
• Molecular wt., pH, isoelectric point are
simple calculations from amino acid
composition
• Hydrophobicity is a property of groups of
amino acids - best examined as a graph
Hydrophobicity Plot
P53_HUMAN (P04637) human cellular tumor antigen p53
Kyte-Doolittle hydrophilicty, window=19
Web Sites for Simple Protein Analysis
• Protein Hydrophobicity Server: Bioinformatics Unit,
Weizmann Institute of Science , Israel
http://bioinformatics.weizmann.ac.il/hydroph/
• SAPS - statistical analysis of protein sequences:
composition, charge, hydrophobic and
transmembrane segments, cysteine spacings,
repeats and periodicity
http://www.isrec.isb-sib.ch/software/SAPS_form.html
EMBOSS Protein Analysis Toolkit
• plotorf: simple open reading frame finder
•
•
•
•
Garnier: predicts 2ndary structure
Charge: plot of protein charge
Octanol: hydrophobicity plot
Pepwindow: hydropathy plot
• pepinfo: plots protein secondary structure and
•
•
•
•
•
hydrophobicity in parallel panels
tmap: predict transmembrane regions
Topo: draws a map of transmembrane protein
Pepwheel: shows protein sequence as helical wheel
Pepcoil: predicts coiled-coil domains
Helixturnhelix: predicts helix-turn-helix domains
Simple Motifs
Common structural motifs
– Membrane spanning
– Signal peptide
– Coiled coil
– Helix-turn-helix
Protein Signal Peptides
• Proteins are sorted
within the cell
using 20-25 amino
acid tags at their 5'
end (beginning)
• Chopped off once
they reach their
destination
Protein Signal Prediction
• ChloroP - Prediction of chloroplast transit peptides
• LipoP - Prediction of lipoproteins and signal peptides in Gram
negative bacteria
• MITOPROT - Prediction of mitochondrial targeting sequences
• PATS - Prediction of apicoplast targeted sequences
• PlasMit - Prediction of mitochondrial transit peptides in Plasmodium
falciparum
• Predotar - Prediction of mitochondrial and plastid targeting
sequences
• PTS1 - Prediction of peroxisomal targeting signal 1 containing
proteins
• SignalP - Prediction of signal peptide cleavage sites･
“Super-secondary” Structure
Common structural motifs
–
–
–
–
Membrane spanning (EMBOSS: tmap, topo)
Signal peptide (EMBOSS: sigcleave)
Coiled coil (EMBOSS: pepcoil)
Helix-turn-helix (EMBOSS: helixturnhelix)
• Predicted from abundance of specific amino
acids in a window and patterns of
hydrophobic/hydrophillic
Web servers that predict
these structures
Predict Protein server: : EMBL Heidelberg
– http://www.embl-heidelberg.de/predictprotein/
SOSUI: Tokyo Univ. of Ag. & Tech., Japan
– http://www.tuat.ac.jp/~mitaku/adv_sosui/submit.html
TMpred (transmembrane prediction): ISREC (Swiss Institute
for Experimental Cancer Research)
– http://www.isrec.isb-sib.ch/software/TMPRED_form.html
COILS (coiled coil prediction): ISREC
– http://www.isrec.isb-sib.ch/software/COILS_form.html
SignalP (signal peptides): Tech. Univ. of Denmark
– http://www.cbs.dtu.dk/services/SignalP/
Protein Domains/Motifs
• Proteins are built out of functional units
know as domains (or motifs)
• These domains have conserved sequences
•
•
Often much more similar than their respective proteins
Exon splicing theory (W. Gilbert)
• Exons correspond to folding domains which in turn
serve as functional units
• Unrelated proteins may share a single similar exon
(i.e.. ATPase or DNA binding function)
Protein Domains
(Pattern analysis)
Motifs are built from
Multiple Alignmennts
Protein Motif Databases
• Known protein motifs have been collected in
databases
• Best database is PROSITE
– The Dictionary of Protein Sites and Patterns
– maintained by Amos Bairoch, at the Univ. of Geneva,
Switzerland
– contains a comprehensive list of documented protein
domains constructed by expert molecular biologists
– Alignments and patterns built by hand!
PROSITE is based on Patterns
Each domain is defined by a simple pattern
– Patterns can have alternate amino acids in each
position and defined spaces, but no gaps
– Pattern searching is by exact matching, so any
new variant will not be found (can allow
mismatches, but this weakens the algorithm)
ID CBD_FUNGAL; PATTERN.
AC PS00562;
DT DEC-1991 (CREATED); NOV-1997 (DATA UPDATE); JUL-1998 (UPDATE).
DE Cellulose-binding domain, fungal type.
PA C-G-G-x(4,7)-G-x(3)-C-x(5)-C-x(3,5)-[NHG]-x-[FYWM]-x(2)-Q-C
Tools for Pattern searching
EMBOSS
fuzznuc: DNA pattern search
fuzzpro: protein pattern search
preg: regular expression search of a
protein sequence
Tools for PROSITE searches
Free Mac program: MacPattern
– ftp://ftp.ebi.ac.uk/pub/software/mac/macpattern.hqx
Free PC program (DOS): PATMAT
– ftp://ncbi.nlm.nih.gov/repository/blocks/patmat.dos
EMBOSS has the programs: patmatdb,
patmatmotifs
Also in virtually all commercial programs: MacVector,
VectorNTI, CLC-Bio, LaserGene, etc.
Websites for PROSITE
Searches
ScanProsite at ExPASy: Univ. of Geneva
– http://expasy.hcuge.ch/sprot/scnpsit1.html
Network Protein Sequence Analysis: Institut de
Biologie et Chimie des Protéines, Lyon, France
– http://pbil.ibcp.fr/NPSA/npsa_prosite.html
PPSRCH: EBI, Cambridge, UK
– http://www2.ebi.ac.uk/ppsearch/
Pattern Search Methods
Complexity
Consensus
exact match
fuzzy match
Pattern
regular expression
(defined mismatches)
PSSM
HMM
Scores for each type of
match in each position,
gapped alignment
Position-specific
gap scores
Challenges to define statistical significance, sensitivity, & specificty
What are all the true postives, & false negatives in a
genome-wide search?
Profiles
• Profiles are tables of amino acid frequencies
at each position in a motif
• They are built from multiple alignments
• PROSITE entries also contain profiles built
from an alignment of proteins that match the
pattern
• Profile searching is more sensitive than
pattern searching - uses an alignment
algorithm, allows gaps
Protein PSSM with log ratios
Profile Alignment
Gribskov et al. 1987
•
•
•
•
Position specific scores
Allows addition of extra sequence(s) to an alignment
Allows alignment of alignments
Gaps introduced as whole columns in the separate
alignments
• Optimal alignment in time O(a2l2)
a = alphabet size, l = sequence length
• Information about the degree of conservation of
sequence positions is included (similar amino acids)
Good reasons to use profile
alignments
– Adding a new sequence to an existing multiple
alignment that you want to keep fixed
(align sequence to profile)
– Searching a database for new members of your
protein family (pfsearch)
– Searching a database of profiles to find out which
one your sequence belongs to (pfscan)
– Combining two multiple sequence alignments
(profile to profile)
EMBOSS ProfileSearch
• EMBOSS has a set of profile analysis tools.
• Start with a multiple alignment
– prophecy: create a profile
– profit: scans a database with your profile
– prophet makes pairwise alignments between a single
sequence and a profile
Websites for Profile searching
• PROSITE ProfileScan: ExPASy, Geneva
– http://www.isrec.isb-sib.ch/software/PFSCAN_form.html
• BLOCKS (builds profiles from PROSITE entries and
adds all matching sequences in SwissProt): Fred Hutchinson
Cancer Research Center, Seattle, Washington, USA
– http://www.blocks.fhcrc.org/blocks_search.html
• PRINTS (profiles built from automatic alignments of
OWL non-redundant protein databases):
http://www.biochem.ucl.ac.uk/cgi-bin/fingerPRINTScan/fps/PathForm.cgi
More Protein Motif Databases
• PFAM (1344 protein family HMM profiles built by
hand): Washington Univ., St. Louis
– http://pfam.wustl.edu/hmmsearch.shtml
• ProDom (profiles built from PSI-BLAST automatic
multiple alignments of the SwissProt database): INRA,
Toulouse, France
– http://www.toulouse.inra.fr/prodom/doc/blast_form.html
[This is my favorite protein database - nicely colored results]
Sample ProDom Output
Profile searching using
PSI-BLAST
• Position Specific Iterative
• Perform search – construct profile – perform
search
• Convergence (hopefully…)
• Increased sensitivity for distantly related
sequences
• Only as good as your first set of hits
• Available on-line (NCBI)
Probabilistic Models of
Sequence Alignment
• Hidden Markov Models
– sequence of states and associated symbol probabilities
• Produces a probabilistic model of a sequence
alignment
• Align a sequence to a Profile Hidden Markov
Model
– Algorithms exist to find the most efficient pathway
through the model
Markov Chain: A sequence of ‘things’. The
probability of the next thing depends only
on the current thing. Based on finite state
automata.
Hidden Markov Model: A sequence of states
which form a Markov Chain. The states are
not observable. The observable characters
have “emission” probabilities which depend
on the current state.
Hidden Markov Models
• Hidden Markov Models (HMMs) are a more
sophisticated form of profile analysis.
• Rather than build a table of amino acid frequencies
at each position, they model the transition from one
amino acid to the next, as well as gaps.
• Pfam is built with HMMs.
• Free HMM software HMMER
• HMMs can be used for a wide range of
bioinformatics problems, not just alignment motifs.
Profile HMM
• The sequence at each position is a “hidden state.” The model contains
probabilities of transitions between states. The “M” box is a Match, which
is further modeled by probabilities for each possible amino acid. There is a
specific probability for Insertion “I” and Deletion “D” at each transition.
• Any sequence can be matched to this model, and its best probability
calculated. The log-odds score is a measure of probability of a sequence
being emitted by an HMM rather than any random (null) model.
Eddy, Sean R., HMMER User Guide, Version 2.3.2; Oct 2003. http://hmmer.wustl.edu/.
Discovery of new Motifs
• All of the tools discussed so far rely on a
database of existing domains/motifs
• How to discover new motifs
–
–
–
–
Start with a set of related proteins
Make a multiple alignment
Build a pattern or profile
You will need access to a fairly powerful UNIX
computer to search databases with custom built
profiles or HMMs.
Patterns in Unaligned
Sequences
• Sometimes sequences may share just a
small common region
–transcription factors
• MEME: San Diego Supercomputing Facility
http://www.sdsc.edu/MEME/meme/website/meme.html
• Gibbs Sampler
• Sombrero (Self-organizing maps)
MEME Details
•
•
•
•
•
The E-value of a motif is based on its log likelihood ratio, width, sites, the background
letter frequencies and the size of the training set. The E-value is an estimate of the
expected number of motifs with the given log likelihood ratio (or higher), and with the
same width and site count, that one would find in a similarly sized set of random
sequences.
Each motif describes a pattern of a fixed width as no gaps are allowed in MEME motifs
log likelihood ratio is the logarithm of the ratio of the probability of the occurrences of the
motif given the motif model (likelihood given the motif) versus their probability given the
background model (likelihood given the null model). (Normally the background model is
a 0-order Markov model using the background letter frequencies, but higher order Markov
models may be specified via the -bfile option to MEME.)
The information content of the motif in bits. It is equal to the sum of the uncorrected
information content, R(), in the columns of the LOGO. This is equal relative entropy of
the motif relative to a uniform background frequency model.
Relative Entropy The relative entropy of the motif, computed in bits and relative to the
background letter frequencies. It is equal to the log-likelihood ratio (llr) divided by the
number of contributing sites of the motif times 1/ln(2),
re = llr / (sites * ln(2)).
True significance of Motifs?
• All motif sampling methods will find common
words in a set of sequences.
• This is essentially a “least common
denominator” approach.
• All sets of biological sequences have some words
above random frequencies.
• Need to compare to an appropriate background
model for motif finding.
• Test found motifs against appropriate positive
and negative controls (how to define?)
Summary
•
•
•
•
•
Restriction sites
Finding genes in DNA sequences
Regulatory sites in DNA
Protein signals (transport and processing)
Protein functional domains & motif
databases
• Regular Expressions
• Position Specific Scoring Matrix
& Hidden Markov Models
0
A:
C:
G:
T:
1
-0.19
1.25
-1.42
-1.42
2
1.46
-1.42
-1.00
-1.42
3
-1.42
1.52
-1.42
-1.42
4
-1.42
-1.42
1.52
-1.42
5
-1.42
-1.42
-1.42
1.52
-1.42
-1.42
1.52
-1.42
Next Lecture: Phylogenetics