BCB 444544

1
BCB 444/544

• Lecture 28
• Gene Prediction - finish it
• Promoter Prediction
• 28_Oct29

2
Required Reading (before lecture)

• Mon Oct 29 - Lecture 28
• Promoter Regulatory Element Prediction
• Chp 9 - pp 113 - 126
• Wed Oct 30 - Lecture 29
• Phylogenetics Basics
• Chp 10 - pp 127 - 141
• Thurs Oct 31 - Lab 9
• Gene Regulatory Element Prediction
• Fri Oct 30 - Lecture 29
• Phylogenetic Tree Construction Methods
  Programs
• Chp 11 - pp 142 - 169

3
Assignments Announcements

• Mon Oct 29 - HW5 - will be posted today
• HW5 Hands-on exercises with
  phylogenetics
• and tree-building software
• Due Mon Nov 5 (not Fri Nov 1 as previously
  posted)

4
BCB 544 "Team" Projects

• Last week of classes will be devoted to Projects
• Written reports due
• Mon Dec 3 (no class that day)
• Oral presentations (20-30') will be
• Wed-Fri Dec 5,6,7
• 1 or 2 teams will present during each class
  period
• See Guidelines for Projects posted online

5
BCB 544 Only New Homework Assignment

• 544 Extra2
• Due vPART 1 - ASAP
• PART 2 - meeting prior to 5 PM Fri Nov 2
• Part 1 - Brief outline of Project, email to Drena
  Michael
• after response/approval, then
• Part 2 - More detailed outline of project
• Read a few papers and summarize status of
  problem
• Schedule meeting with Drena Michael to
  discuss ideas

6
Seminars this Week

• BCB List of URLs for Seminars related to
  Bioinformatics
• http//www.bcb.iastate.edu/seminars/index.html
• Nov 1 Thurs - BBMB Seminar 410 in 1414 MBB
• Todd Yeates UCLA TBA -something cool about
  structure and evolution?
• Nov 2 Fri - BCB Faculty Seminar 210 in 102 ScI
• Bob Jernigan BBMB, ISU
• Control of Protein Motions by Structure

7
Chp 8 - Gene Prediction

• SECTION III GENE AND PROMOTER PREDICTION
• Xiong Chp 8 Gene Prediction
• Categories of Gene Prediction Programs
• Gene Prediction in Prokaryotes
• Gene Prediction in Eukaryotes

8
Computational Gene Prediction Approaches

• Ab initio methods
• Search by signal find DNA sequences involved in
  gene expression
• Search by content Test statistical properties
  distinguishing coding from non-coding DNA
• Similarity-based methods
• Database search exploit similarity to proteins,
  ESTs, cDNAs
• Comparative genomics exploit aligned genomes
• Do other organisms have similar sequence?
• Hybrid methods - best

9
Computational Gene Prediction Algorithms
This is a new slide

• Neural Networks (NNs) (more on these later)
• e.g., GRAIL
• Linear discriminant analysis (LDA) (see text)
• e.g., FGENES, MZEF
• Markov Models (MMs) Hidden Markov Models (HMMs)
  
• e.g., GeneSeqer - uses MMs
• GENSCAN - uses 5th order HMMs - (see
  text)
• HMMgene - uses conditional maximum
  likelihood (see text)

10
Signals Search
This is a new slide

• Approach Build models (PSSMs, profiles, HMMs,
  ) and search against DNA. Detected instances
  provide evidence for genes

11
Content Search
This is a new slide

• Observation Encoding a protein affects
  statistical properties of DNA sequence
• Nucleotide.amino acid distribution
• GC content (CpG islands, exon/intron)
• Uneven usage of synonymous codons (codon bias)
• Hexamer frequency - most discriminative of these
  for identifying coding potential
• Method Evaluate these differences (coding
  statistics) to differentiate between coding and
  non-coding regions

12
Human Codon Usage
This is a new slide
13
Predicting Genes based on Codon Usage
Differences
This is a new slide

• Algorithm
• Process sliding window
• Use codon frequencies to compute probability of
  coding versus non-coding
• Plot log-likelihood ratio

14
Similarity-Based Methods Database Search
This is a new slide

• In different genomes Translate DNA into all 6
  reading frames and search against proteins
  (TBLASTX,BLASTX, etc.)
• Within same genome Search with EST/cDNA database
  
• (EST2genome, BLAT, etc.).
• Problems
• Will not find new or RNA genes (non-coding
  genes).
• Limits of similarity are hard to define
• Small exons might be overlooked

15
Similarity-Based Methods Comparative Genomics
This is a new slide

• Idea Functional regions are more conserved than
  non-functional ones high similarity in alignment
  indicates gene
• Advantages
• May find uncharacterized or RNA genes
• Problems
• Finding suitable evolutionary distance
• Finding limits of high similarity (functional
  regions)

16
This is a new slide
Human-Mouse Homology

• Comparison of 1196 orthologous genes
• Sequence identity between genes in human vs mouse
• Exons 84.6
• Protein 85.4
• Introns 35
• 5 UTRs 67
• 3 UTRs 69

17
Thanks to Volker Brendel, ISU for the following
Figs Slides

• Slightly modified from
• BSSI Genome Informatics Module
• http//www.bioinformatics.iastate.edu/BBSI/course_
  desc_2005.htmlmoduleB
• V Brendel vbrendel_at_iastate.edu

Brendel et al (2004) Bioinformatics 20 1157
18
Spliced Alignment Algorithm
GeneSeqer - Brendel et al.- ISU
http//deepc2.psi.iastate.edu/cgi-bin/gs.cgi
Brendel et al (2004) Bioinformatics 20
1157 http//bioinformatics.oxfordjournals.org/cgi/
content/abstract/20/7/1157

• Perform pairwise alignment with large gaps in one
  sequence (due to introns)
• Align genomic DNA with cDNA, ESTs, protein
  sequences
• Score semi-conserved sequences at splice
  junctions
• Using Bayesian probability model 1st order MM
• Score coding constraints in translated exons
• Using Bayesian model

Brendel 2005
19
Splice Site Detection
Do DNA sequences surrounding splice "consensus"
sequences contribute to splicing signal?
YES
i ith position in sequence I avg
information content over all positions gt20 nt
from splice site ?I avg sample standard
deviation of I
Brendel 2005
20
Information Content vs Position
Which sequences are exons which are introns?
How can you tell?
Brendel 2005
21
Markov Model for Spliced Alignment
Brendel 2005
22
This is a new slide
Evaluation of Splice Site Prediction
TP positive instance correctly predicted as
positive FP negative instance incorrectly
predicted as positive TN negative instance
correctly predicted as negative FN positive
instance incorrectly predicted as negative
Right!
Fig 5.11 Baxevanis Ouellette 2005
23
Evaluation of Predictions
Predicted Positives
True Positives
False Positives
Coverage
Recall
Do not memorize this!
24
Evaluation of Predictions - in English
Coverage
IMPORTANT Sensitivity alone does not tell us
much about performance because a 100 sensitivity
can be achieved trivially by labeling all test
cases positive!
In English? Sensitivity is the fraction of all
positive instances having a true positive
prediction.
Recall
IMPORTANT in medical jargon, Specificity is
sometimes defined differently (what we define
here as "Specificity" is sometimes referred to as
"Positive predictive value")
In English? Specificity is the fraction of all
predicted positives that are, in fact, true
positives.
25
This slide has been changed
Best Measures for Comparison?

• ROC curves (Receiver Operating Characteristic
  (?!!)
• http//en.wikipedia.org/wiki/Roc_curve
• Correlation Coefficient
• Matthews correlation coefficient (MCC)
• MCC 1 for a perfect prediction
• 0 for a completely random assignment
• -1 for a "perfectly incorrect" prediction

In signal detection theory, a receiver operating
characteristic (ROC), or ROC curve is a plot of
sensitivity vs (1 - specificity) for a binary
classifier system as its discrimination threshold
is varied. The ROC can also be represented
equivalently by plotting fraction of true
positives (TPR true positive rate) vs fraction
of false positives (FPR false positive rate)
Do not memorize this!
26
GeneSeqer Input http//deepc2.psi.iastate.edu/cg
i-bin/gs.cgi
Brendel 2005
27
GeneSeqer Output
Brendel 2005
28
GeneSeqer Gene Evidence Summary
Brendel 2005
29
Gene Prediction - Problems Status?

• Common errors?
• False positive intergenic regions
• 2 annotated genes actually correspond to a single
  gene
• False negative intergenic region
• One annotated gene structure actually contains 2
  genes
• False negative gene prediction
• Missing gene (no annotation)
• Other
• Partially incorrect gene annotation
• Missing annotation of alternative transcripts
• Current status?
• For ab initio prediction in eukaryotes HMMs have
  better overall performance for detecting
  intron/exon boundaries
• Limitation? Training data predictions are
  organism specific
• Combined ab initio/homology based predictions
  Improved accurracy
• Limitation? Availability of identifiable
  sequence homologs in databases

30
Recommended Gene Prediction Software

• Ab initio
• GENSCAN http//genes.mit.edu/GENSCAN.html
• GeneMark.hmm http//exon.gatech.edu/GeneMark/
• others GRAIL, FGENES, MZEF, HMMgene
• Similarity-based
• BLAST, GenomeScan, EST2Genome, Twinscan
• Combined
• GeneSeqer, http//deepc2.psi.iastate.edu/cgi-bin/g
  s.cgi
• ROSETTA
• Consensus because results depend on organisms
  specific task, Always use more than one
  program!
• Two servers hat report consensus predictions
• GeneComber
• DIGIT

31
Other Gene Prediction Resources at ISU
http//www.bioinformatics.iastate.edu/bioinformati
cs2go/
32
Other Gene Prediction Resources GaTech, MIT,
Stanford, etc.
Lists of Gene Prediction Software http//www.bioi
nformaticsonline.org/links/ch_09_t_1.html http//
cmgm.stanford.edu/classes/genefind/

• Current Protocols in Bioinformatics (BCB/ISU owns
  a copy - currently in my lab!)
• Chapter 4 Finding Genes
• 4.1 An Overview of Gene Identification
  Approaches, Strategies, and Considerations
• 4.2 Using MZEF To Find Internal Coding Exons
• 4.3 Using GENEID to Identify Genes
• 4.4 Using GlimmerM to Find Genes in Eukaryotic
  Genomes
• 4.5 Prokaryotic Gene Prediction Using GeneMark
  and GeneMark.hmm
• 4.6 Eukaryotic Gene Prediction Using GeneMark.hmm
  
• 4.7 Application of FirstEF to Find Promoters and
  First Exons in the Human Genome
• 4.8 Using TWINSCAN to Predict Gene Structures in
  Genomic DNA Sequences
• 4.9 GrailEXP and Genome Analysis Pipeline for
  Genome Annotation
• 4.10 Using RepeatMasker to Identify Repetitive
  Elements in Genomic Sequences

33
Chp 9 - Promoter Regulatory Element Prediction

• SECTION III GENE AND PROMOTER PREDICTION
• Xiong Chp 9 Promoter Regulatory Element
  Prediction
• Promoter Regulatory Elements in Prokaryotes
• Promoter Regulatory Elements in Eukaryotes
• Prediction Algorithms

34
Eukaryotes vs Prokaryotes Genomes

• Eukaryotic genomes
• Are packaged in chromatin sequestered in a
  nucleus
• Are larger and have multiple linear chromosomes
• Contain mostly non-protein coding DNA (98-99)
• Prokarytic genomes
• DNA is associated with a nucleoid, but no nucleus
• Much larger, usually single, circular chromosome
• Contain mostly protein encoding DNA

35
Eukaryotes vs Prokryotes Gene Structure
36
Eukaryotes vs Prokaryotes Genes

• Eukaryotic genes
• Are larger and more complex than in prokaryotes
• Contain introns that are spliced out to
  generate mature mRNAs
• Often undergo alternative splicing, giving rise
  to multiple RNAs
• Are transcribed by 3 different RNA polymerases
• (instead of 1, as in prokaryotes)
• In biology, statements such as this include an
  implicit usually or often

37
Eukaryotes vs Prokaryotes Levels of Gene
Regulation

• Primary level of control?
• Prokaryotes Transcription initiation
• Eukaryotes Transcription is also very
  important, but
• Expression is regulated at multiple levels
• many of which are post-transcriptional
• RNA processing, transport, stability
• Translation initiation
• Protein processing, transport, stability
• Post-translational modification (PTM)
• Subcellular localization
• Recent important discoveries small regulatory
  RNAs (miRNA, siRNA) are abundant and play very
  important roles in controlling gene expression in
  eukaryotes, often at post-transcriptional levels

38
Eukaryotes vs Prokaryotes Regulatory Elements

• Prokaryotes
• Promoters operators (for operons) - cis-acting
  DNA signals
• Activators repressors - trans-acting proteins
• (we won't discuss these)
• Eukaryotes
• Promoters enhancers (for single genes) -
  cis-acting
• Transcription factors - trans-acting
• Important difference?
• What the RNA polymerase actually binds

39
Prokaryotic Promoters

• RNA polymerase complex recognizes promoter
  sequences located very close to and on 5 side
  (upstream) of tansription initiation site
• Prokaryotic RNA polymerase complex binds directly
  to promoter, by virtue of its sigma subunit - no
  requirement for transcription factors binding
  first
• Prokaryotic promoter sequences are highly
  conserved
• -10 region
• -35 region

40
Eukaryotic Promoters

• Eukaryotic RNA polymerase complexes do not bind
  directly to promoter sequences
• Transcription factors must bind first and serve
  as landmarks recognized by RNA polymerase
  complexes
• Eukaryotic promoter sequences are less highly
  conserved, but many promoters (for RNA
  polymerase II) contain
• -30 region "TATA" box
• -100 region "CCAAT" box

41
Eukaryotic Promoters vs Enhancers

• Both promoters enhancers are binding sites for
  transcription factors (TFs)
• Promoters
• essential for initiation of transcription
• located relatively close to start site (usually
  lt200 bp upstream, but can be located within
  gene, rather than upstream!)
• Enhancers
• needed for regulated transcription (differential
  expression in specific cell types, developmental
  stages, in response to environment, etc.)
• can be very far from start site (sometimes gt 100
  kb)

42
Eukaryotic genes are transcribed by 3 different
RNA polymerases (Location of promoter regions,
TFBSs TFs differ, too)
Brown Fig 9.18
BIOS Scientific Publishers Ltd, 1999
43
Prokaryotic Genes Operons

• Genes with related functions are often clustered
  within operons (e.g., lac operon)
• Operons genes with related functions that are
  transcribed and regulated as a single unit one
  promoter controls expression of several proteins
• mRNAs produced from operons are polycistronic -
  a single mRNA encodes several proteins i.e.,
  there are multiple ORFs, each with its own AUG
  (START) STOP codons, linked within one mRNA
  molecule

44
Promoter of lac operon in E. coli (Transcribed
by prokaryotic RNA polymerase)
Brown Fig 9.17
BIOS Scientific Publishers Ltd, 1999
45
Eukaryotic genes

• Genes with related functions are occasionally,
  but not usually clustered instead, they share
  common regulatory regions (promoters, enhancers,
  etc.)
• Chromatin structure must also be active for
  transcription to occur

46
Eukaryotic genes have large complex
regulatory regions

• Cis-acting regulatory elements include
• Promoters, enhancers, silencers
• Trans-acting regulatory factors include
• Transcription factors (TFs), chromatin
• remodeling complexes, small RNAs

Brown Fig 9.17
BIOS Scientific Publishers Ltd, 1999
47
Eukaryotic Promoters DNA sequences required for
initiation, usually lt200 bp from start site
Eukaryotic RNA polymerases bind by recognizing a
complex of TFs bound at promotor
First, TFs must bind short motifs (TFBSs) within
promoters then RNA polymerase can bind and
initiate transcription of RNA
250 bp
Pre-mRNA
48
Eukaryotic promoters enhancer regions often
contain many different TFBS motifs
Fig 9.13 Mount 2004
49
Simplified View of Promoters in Eukaryotes
Fig 5.12 Baxevanis Ouellette 2005
50
Eukaryotic Activators vs Repressors
Regions far from the promoter can act as
"enhancers" or "repressors" of transcription by
serving as binding sites for activator or
repressor proteins (TFs)
Activator proteins (TFs) bind to enhancers
interact with RNAP to stimulate transcription
Repressors block the action of activators
51
Eukaryotic Transcription Factors (TFs)

• Transcription factors proteins that interact
  with the RNA polymerase complex to activate or
  repress transcription
• TFs often contain both
• a trans-activating domain
• a DNA binding domain or motif
• TFs recognize and bind specific short DNA
  sequence motifs called transcription factor
  binding sites (TFBSs)
• Databases for TFs TFBSs include
• TRANSFAC, http//www.generegulation.com/cgibin/pub
  /databases/transfac
• JASPAR

Here motif amino acid sequence in protein
Here motif nucleotide sequence in DNA
52
Zinc Finger Proteins - Transcription Factors

• Common in eukaryotic proteins
• 1 of mammalian genes encode zinc-finger
  proteins (ZFPs)
• In C. elegans, there are gt 500 !
• Can be used as highly specific DNA binding
  modules
• Potentially valuable tools for directed genome
  modification (esp. in plants) human gene
  therapy - one clinical trial will begin soon!

• Did you go to Dave Segal's seminar?
• Your TAs Pete Jeff work on designing better
  ZFPs!

Brown Fig 9.12
BIOS Scientific Publishers Ltd, 1999
53
Promoter Prediction Algorithms Software
Xiong -
54
Eukaryotes vs Prokaryotes Promoter Prediction
Promoter prediction is much easier in
prokaryotes Why? Highly conserved Simpler
gene structures More sequenced genomes!
(for comparative approaches) Methods?
Previously mostly HMM-based Now
similarity-based comparative methods because
so many genomes available Xiong textbook 1)
"Manual method" rules of Wang et al (see
text) 2) BPROM - uses linear discriminant
function
55
Eukaryotes vs Prokaryotes Promoter Prediction
Promoter prediction is much easier in
prokaryotes Why? Highly conserved Simpler
gene structures More sequenced genomes!
(for comparative approaches) Methods?
Previously mostly HMM-based Now
similarity-based comparative methods because
so many genomes available Xiong textbook 1)
"Manual method" rules of Wang et al (see
text) 2) BPROM - uses linear discriminant
function
56
Predicting Promoters in Eukaryotes

• Closely related to gene prediction!
• Obtain genomic sequence
• Use sequence-similarity based comparison
• (BLAST, MSA) to find related genes
• But "regulatory" regions are much less
  well- conserved than coding regions
• Locate ORFs
• Identify Transcription Start Site (TSS)
• (if possible!)
• Use Promoter Prediction Programs
• Analyze motifs, etc. in DNA sequence (TRANSFAC,
  JASPAR)

57
Predicting promoters Steps Strategies

• Identify TSS --if possible?
• One of biggest problems is determining exact
  TSS!
• Not very many full-length cDNAs!
• Good starting point? (human vertebrate genes)
• Use FirstEF
• found within UCSC Genome Browser
• or submit to FirstEF web server

Fig 5.10 Baxevanis Ouellette 2005
58
Automated Promoter Prediction Strategies

• Pattern-driven algorithms (ab initio)
• Sequence-driven algorithms (homology based)
• Combined "evidence-based"
• BEST RESULTS? Combined, sequential

59
1) Pattern-driven Algorithms

• Success depends on availability of collections of
  annotated transcription factor binding sites
  (TFBSs)
• Tend to produce very large numbers of false
  positives (FPs)
• Why?
• Binding sites for specific TFs are often variable
• Binding sites are short (typically 6-10 bp)
• Interactions between TFs ( other proteins)
  influence both affinity specificity of TF
  binding
• One binding site often recognized by multiple TFs
  
• Biology is complex gene activation is often
  specific to organism/cell/stage/environmental
  condition promoter and enhancer elements must
  mediate this

60
Ways to Reduce FPs in ab initio Prediction

• Take sequence context/biology into account
• Eukaryotes clusters of TFBSs are common
• Prokaryotes knowledge of ? (sigma) factors helps
• Probability of "real" binding site higher if
  annotated transcription start site (TSS) is
  nearby
• But What about enhancers? (no TSS nearby!)
• only a small fraction of TSSs have been
  experimentally determinined
• Do the wet lab experiments!
• But Promoter-bashing can be tedious

61
2) Sequence-driven Algorithms

• Assumption Common functionality can be deduced
  from sequence conservation (Homology)
• Alignments of co-regulated genes should highlight
  elements involved in regulation
• Careful How determine co-regulation?
• Orthologous genes from difference species
• Genes experimentally shown to be co-regulated
  (using microarrays??)
• Comparative promoter prediction
• Phylogenetic footprinting
• Expression Profiling

62
Phylogenetic Footprinting

• Based on increasing availability of whole genome
  DNA sequences from many different species
• Selection of organisms for comparison is
  important
• not too close, not too far good human vs
  mouse
• To reduce FPs, must extract non-coding sequences
  and then align them prediction depends on good
  alignment
• use MSA algorithms (e.g., CLUSTAL)
• more sensitive methods
• Gibbs sampling
• Expectation Maximization (EM) methods
• Examples of programs
• Consite, rVISTA, PromH(W), Bayes aligner,
  Footprinter

63
Expression Profiling

• Based on increasing availability of whole genome
  mRNA expression data, esp., microarray data
• High-throughput simultaneous monitoring of
  expression levels of thousands of genes
• Assumptions (sometimes valid, sometimes NOT)
• Co-expression implies co-regulation
• Co-regulated genes share common regulatory
  elements
• Drawbacks
• Signals are short weak!
• Requires Gibbs sampling or EM e.g.,
  MEME, AlignACE, Melina
• Prediction depends on determining which genes are
  co-expressed - usually by clustering -
  which an be error prone
• Examples of programs
• INCLUSive - combined microarray analysis motif
  detection
• PhyloCon - combined phylo footprinting
  expression profiling)

64
Problems with Sequence-driven Algorithms

• Need sets of co-regulated genes
• For comparative (phylogenetic) methods
• Must choose appropriate species
• Different genomes evolve at different rates
• Classical alignment methods have trouble with
• translocations or inversions than change
  order of functional elements
• If background conservation of entire region is
  high, comparison is useless
• Not enough data (but Prokaryotes gtgtgt Eukaryotes)
• Complexity many regulatory elements are not
  conserved across species!

65
TRANSFAC Matrix Entry for TATA box

• Fields
• Accession ID
• Brief description
• TFs associated with this entry
• Weight matrix
• Number of sites used to build
• Other info

Fig 5.13 Baxevanis Ouellette 2005
66
Global Alignment of Human Mouse Obese Gene
Promoters (200 bp upstream from TSS)
Fig 5.14 Baxevanis Ouellette 2005
67
Annotated Lists of Promoter Databases Promoter
Prediction Software

• URLs from Mount textbook
• Table 9.12 http//www.bioinformaticsonline.org/li
  nks/ch_09_t_2.html
• Table in Wasserman Sandelin Nat Rev Genet
  article http//proxy.lib.iastate.edu2103/nrg/jour
  nal/v5/n4/full/nrg1315_fs.htm
• URLs from Baxevanis Ouellette textbook
• http//www.wiley.com/legacy/products/subject/life
  /bioinformatics/ch05.htmlinks
• More lists
• http//www.softberry.com/berry.phtml?topicindexg
  roupprogramssubgrouppromoter
• http//bioinformatics.ubc.ca/resources/links_direc
  tory/?subcategory_id104
• http//www3.oup.co.uk/nar/database/subcat/1/4/

68
Check out Optional Review Try Associated
Tutorial

• Wasserman WW Sandelin A (2004) Applied
  bioinformatics for identification of regulatory
  elements. Nat Rev Genet 5276-287
• http//proxy.lib.iastate.edu2103/nrg/journal/v5/
  n4/full/nrg1315_fs.html

Check this out http//www.phylofoot.org/NRG_test
cases/
Bottom line this is a very "hot" area - new
software for computational prediction of gene
regulatory elements published every day!