copyright © 2004 by limsoon wong gene finding & gene feature recognition by computational...

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Gene Finding & Gene Feature

Recognition by Computational

Analysis

Limsoon WongInstitute for Infocomm

ResearchNovember 2004

Copyright © 2004 by Limsoon Wong

Lecture Plan

• Gene structure basics• Gene finding overview• GRAIL• Indel & frame-shift in coding regions• Histone promoters: A cautionary case

study• Knowledge discovery basics• TIS recognition• Poly-A signal recognition• TSS recognition

Basic materials

Advanced materials

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Gene Structure Basics

A brief refresher

Some slides here are “borrowed” from Ken Sung


Body

• Our body consists of a number of organs• Each organ composes of a number of

tissues• Each tissue composes of cells of the

same type


Cell

• Performs two types of function– Chemical reactions necessary to maintain our

life– Pass info for maintaining life to next

generation

• In particular– Protein performs chemical reactions– DNA stores & passes info– RNA is intermediate between DNA & proteins

Protein

• A protein sequence composed from an alphabet of 20 amino acids– Length is usually 20 to

5000 amino acids – Average around 350

amino acids

• Folds into 3D shape, forming the building blocks & performing most of the chemical reactions within a cell


Amino Acid

• Each amino acid consist of– Amino group– Carboxyl group– R group

Carboxyl group

Aminogroup

C

(the central carbon)R group

NH2

H

C C

R

OH

O



Classification of Amino Acids

• Amino acids can be classified into 4 types.

• Positively charged (basic)– Arginine (Arg, R)– Histidine (His, H)– Lysine (Lys, K)

• Negatively charged (acidic)– Aspartic acid (Asp, D)– Glutamic acid (Glu, E)


Classification of Amino Acids• Polar (overall

uncharged, but uneven charge distribution. can form hydrogen bonds with water. they are called hydrophilic)– Asparagine (Asn, N)– Cysteine (Cys, C)– Glutamine (Gln, Q)– Glycine (Gly, G)– Serine (Ser, S)– Threonine (Thr, T)– Tyrosine (Tyr, Y)

• Nonpolar (overall uncharged and uniform charge distribution. cant form hydrogen bonds with water. they are called hydrophobic)– Alanine (Ala, A)– Isoleucine (Ile, I)– Leucine (Leu, L)– Methionine (Met, M)– Phenylalanine (Phe, F)– Proline (Pro, P)– Tryptophan (Trp, W)– Valine (Val, V)

N

H

C C

R’

OH

O

NH2

H

C C

R

O

H

Peptide bond

NH2

H

C C

R

OH

O

NH2

H

C C

R’

OH

O

+

Protein & Polypeptide Chain

• Formed by joining amino acids via peptide bond• One end the amino group, called N-terminus• The other end is the carboxyl group, called C-

terminus


DNA

• DNA stores instruction needed by the cell to perform daily life function

• Consists of two strands interwoven together and form a double helix

• Each strand is a chain of some small molecules called nucleotides

Francis Crick shows James Watson the model of DNA in their room number 103 of the Austin Wing at the Cavendish Laboratories, Cambridge


Base(Adenine)

DeoxyribosePhosphate

5`

4`3`

2`

1`

Nucleotide

• Consists of three parts:– Deoxyribose– Phosphate (bound to the 5’ carbon)– Base (bound to the 1’ carbon)


A C G T U

Classification of Nucleotides

• 5 diff nucleotides: adenine(A), cytosine(C), guanine(G), thymine(T), & uracil(U)

• A, G are purines. They have a 2-ring structure• C, T, U are pyrimidines. They have a 1-ring

structure• DNA only uses A, C, G, & T


A

T

10Å

G

C

10Å

Watson-Crick rules

• Complementary bases:– A with T (two hydrogen-bonds)– C with G (three hydrogen-bonds)


P P P P

5’3’

A C G T A

Orientation of a DNA

• One strand of DNA is generated by chaining together nucleotides, forming a phosphate-sugar backbone

• It has direction: from 5’ to 3’, because DNA always extends from 3’ end:– Upstream, from 5’ to 3’– Downstream, from 3’ to 5’


Double Stranded DNA

• DNA is double stranded in a cell. The two strands are anti-parallel. One strand is reverse complement of the other

• The double strands are interwoven to form a double helix



Locations of DNAs in a Cell?

• Two types of organisms– Prokaryotes (single-celled organisms with no nuclei. e.g.,

bacteria)

– Eukaryotes (organisms with single or multiple cells. their cells have nuclei. e.g., plant & animal)

• In Prokaryotes, DNA swims within the cell• In Eukaryotes, DNA locates within the

nucleus


Chromosome

• DNA is usually tightly wound around histone proteins and forms a chromosome

• The total info stored in all chromosomes constitutes a genome

• In most multi-cell organisms, every cell contains the same complete set of chromosomes – May have some small different due to

mutation

• Human genome has 3G base pairs, organized in 23 pairs of chromosomes


Gene

• A gene is a sequence of DNA that encodes a protein or an RNA molecule

• About 30,000 – 35,000 (protein-coding) genes in human genome

• For gene that encodes protein– In Prokaryotic genome, one gene corresponds

to one protein– In Eukaryotic genome, one gene can

corresponds to more than one protein because of the process “alternative splicing”


Complexity of Organism vs. Genome Size

• Human Genome: 3G base pairs

• Amoeba dubia (a single cell organism): 600G base pairs

Genome size has no relationship with the complexity of the organism


Number of Genes vs. Genome Size

• Prokaryotic genome (e.g., E. coli)– Number of base pairs: 5M– Number of genes: 4k– Average length of a gene:

1000 bp

• Eukaryotic genome (e.g., human)– Number of base pairs: 3G– Estimated number of

genes: 30k – 35k– Estimated average length

of a gene: 1000-2000 bp

• ~ 90% of E. coli genome are of coding regions.

• < 3% of human genome is believed to be coding regions

Genome size has no relationship with the number of genes!

Base(Adenine)

Ribose SugarPhosphate

5`

4`3`

2`

1`

RNA

• RNA has both the properties of DNA & protein– Similar to DNA, it can

store & transfer info– Similar to protein, it

can form complex 3D structure & perform some functions

• Nucleotide for RNA has of three parts:– Ribose Sugar (has an

extra OH group at 2’)– Phosphate (bound to 5’

carbon)– Base (bound to 1’

carbon)



RNA vs DNA

• RNA is single stranded• Nucleotides of RNA are similar to that of

DNA, except that have an extra OH at position 2’– Due to this extra OH, it can form more

hydrogen bonds than DNA– So RNA can form complex 3D structure

• RNA use the base U instead of T– U is chemically similar to T– In particular, U is also complementary to A

Mutation

• Mutation is a sudden change of genome

• Basis of evolution• Cause of cancer• Can occur in DNA,

RNA, & Protein


Central Dogma

• Gene expression consists of two steps– Transcription

DNA mRNA

– Translation mRNA Protein



Transcription

• Synthesize mRNA from one strand of DNA– An enzyme RNA

polymerase temporarily separates double-stranded DNA

– It begins transcription at transcription start site

– A A, CC, GG, & TU

– Once RNA polymerase reaches transcription stop site, transcription stops

• Additional “steps” for Eukaryotes– Transcription produces

pre-mRNA that contains both introns & exons

– 5’ cap & poly-A tail are added to pre-mRNA

– RNA splicing removes introns & mRNA is made

– mRNA are transported out of nucleus


Translation

• Synthesize protein from mRNA

• Each amino acid is encoded by consecutive seq of 3 nucleotides, called a codon

• The decoding table from codon to amino acid is called genetic code

• 43=64 diff codons Codons are not 1-to-1

corr to 20 amino acids• All organisms use the

same decoding table• Recall that amino acids

can be classified into 4 groups. A single-base change in a codon is usually not sufficient to cause a codon to code for an amino acid in different group

Genetic Code• Start codon: ATG (code for M)• Stop codon: TAA, TAG, TGA



Ribosome

• Translation is handled by a molecular complex, ribosome, which consists of both proteins & ribosomal RNA (rRNA)

• Ribosome reads mRNA & the translation starts at a start codon (the translation start site)

• With help of tRNA, each codon is translated to an amino acid

• Translation stops once ribosome reads a stop codon (the translation stop site)


Introns and exons

• Eukaryotic genes contain introns & exons– Introns are seq that

are ultimately spliced out of mRNA

– Introns normally satisfy GT-AG rule, viz. begin w/ GT & end w/ AG

– Each gene can have many introns & each intron can have thousands bases

• Introns can be very long

• An extreme example is a gene associated with cystic fibrosis in human:– Length of 24 introns

~1Mb– Length of exons ~1kb

• Unlike eukaryotic genes, a prokaryotic gene typically consists of only one contiguous coding region

Typical Eukaryotic Gene Structure


Image credit: Xu

Reading frame #1

ATGGCTTACGCTTGC

Reading frame #2

TGGCTTACGCTTGA.

Reading frame #3

GGCTTACGCTTGA..

ATGGCTTACGCTTGAForward strand:

Reading frame #4

TCAAGCGTAAGCCAT

Reading frame #5

CAAGCGTAAGCCAT.

Reading frame #6

AAGCGTAAGCCAT..

TCAAGCGTAAGCCATReverse strand:

Reading Frame

• Each DNA segment has six possible reading frames


stop stop

ORF

Open Reading Frame (ORF)

• ORF is a segment of DNA with two in-frame stop codons at the two ends and no in-frame stop codon in the middle

• Each ORF has a fixed reading frame



Coding Region

• Each coding region (exon or whole gene) has a fixed translation frame

• A coding region always sits inside an ORF of same reading frame

• All exons of a gene are on the same strand

• Neighboring exons of a gene could have different reading frames

ATG GCT TGG GCT TTA A -------------- GT TTC CCG GAG AT ------ T GGG

exon 1 exon 3exon 2

Frame Consistency

• Neighbouring exons of a gene should be frame-consistent


Exercise:

Define frame consistency mathematically

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Any Question?

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Overview of Gene Finding

Some slides here are “borrowed” from Mark Craven

What is Gene Finding?

• Find all coding regions from a stretch of DNA sequence, and construct gene structures from the identified exons

• Can be decomposed into– Find coding potential

of a region in a frame– Find boundaries betw

coding & non-coding regions


Image credit: Xu


Approaches

• Search-by-signal: find genes by identifying the sequence signals involved in gene expression

• Search-by-content: find genes by statistical properties that distinguish protein coding DNA from non-coding DNA

• Search-by-homology: find genes by homology (after translation) to proteins

• State-of-the-art systems for gene finding usually combine these strategies

Relevant Signals for Search-by-Signals

• Transcription initiation– Promoter

• Transcription termination– Terminators

• Translation initiation– Ribosome binding sites– Initiation codons

• Translation termination– Stop codons

• RNA processing– Splice junction


Image credit: Xu


How Search-by-Signal Works

• There are 2 impt regions in a promoter seq–10 region, ~10bp before TSS–35 region, ~35bp before TSS

• Consensus for–10 region in E. coli is TATAAT, but few promoters actually have this seq

Recognize promoters by– weight matrices– probabilistic models– neural networks, …

How Search-by-Content Works

• Encoding a protein affects stats properties of a DNA seq– some amino acids used

more frequently– diff number of codons

for diff amino acids– for given protein,

usually one codon is used more frequently than others

Estimate prob that a given region of seq was “caused by” its being a coding seq


Image credit: Craven


How Search-by-Homology Works

• Translate DNA seq in all reading frames• Search against protein db• High-scoring matches suggest presence

of homologous genes in DNA You can use BLASTX for this


Search-by-Content Example: Codon Usage Method

• Staden & McLachlan, 1982• Process a seq w/ “window” of length L• Assume seq falls into one of 7 categories,

viz.– Coding in frame 0, frame 1, …, frame 5– Non-coding

• Use Bayes’ rule to determine prob of each category

• Assign seq to category w/ max prob


• Pr(codingi) is the same for each frame if window size fits same number of codons in each frame

• otherwise, consider relative number of codons in window in each frame

Genbank or nr

candidate gene

BLAST search

sequence alignments with known genes,

alignment p-values

Image credit: Xu


Search-by-Homology Example: Gene Finding Using

BLAST• High seq similarity typically implies

homologous genes Search for genes in yeast seq using

BLAST Extract Feature for gene identification

BLAST hits

sequence

• Searching all ORFs against known genes in nr db helps identify an initial set of (possibly incomplete) genes

Image credit: Xu

• A (yeast) gene starts w/ ATG and ends w/ a stop codon, in same reading frame of ORF

• Have “strong” coding potentials, measured by, preference models, Markov chain model, ...

• Have “strong” translation start signal, measured by weight matrix model, ...

• Have distributions wrt length, G+C composition, ...

• Have special seq signals in flanking regions, ...

known genes

0

%known non-

genes

coding potential

gene length distribution

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Any Question?

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

GRAIL,An Important Gene Finding Program

• Signals assoc w/ coding regions • Models for coding regions• Signals assoc w/ boundaries • Models for boundaries• Other factors & information fusion

Some slides here are “borrowed” from Ying Xu

Coding Signal

• Freq distribution of dimers in protein sequence

• E.g., Shewanella– Ave freq is 5%– Some amino

acids prefer to be next to each other

– Some amino acids prefer to be not next to each other


Exercise: What is shewanella?

Image credit: Xu


Coding Signal

• Dimer preference implies dicodon (6-mers like AAA TTT) bias in coding vs non-coding regions

• Relative freq of a di-codon in coding vs non-coding– Freq of dicodon X (e.g, AAA AAA) in coding region, total

number of occurrences of X divided by total number of dicocon occurrences

– Freq of dicodon X (e.g, AAA AAA) in noncoding region, total number of occurrences of X divided by total number of dicodon occurrences

• Exercise: In human genome, freq of dicodon “AAA AAA” is ~1% in coding region vs ~5% in non-coding region. If you see a region with many “AAA AAA”, would you guess it is a coding or non-coding region?


There are

43 = 64 codons

46 = 4096 dicodons

49 = 262144 tricodons

Why Dicodon (6-mer)?

• Codon (3-mer)-based models are not as info rich as dicodon-based models

• Tricodon (9-mer)-based models need too many data points

• To make stats reliable, need ~15 occurrences of each X-mer

For tricodon-based models, need at least 15*262144 = 3932160 coding bases in our training data, which is probably not going to be available for most genomes


• Most dicodons show bias towards either coding or non-coding regions

Foundation for coding region identification

Dicodon freq are key signal used for coding region detection; all gene finding programs use this info

Regions consisting of dicodons that mostly tend to be in coding regions are

probably coding regions; otherwise non-coding regions

Coding Signal

Shewanella Bovine

Coding Signal

• Dicodon freq in coding vs non-coding are genome-dependent


Image credit: Xu

• In-frame vs any-frame dicodons

ATG TTG GAT GCC CAG AAG.....

in-frame dicodons

not in-frame dicodonsIn-frame:

ATG TTGGAT GCCCAG AAG

Not in-frame:

TGTTGG, ATGCCCAGAAG ., GTTGGAAGCCCA, AGAAG ..

any-frame

Coding Signal

• In-frame dicodon freq provide a more sensitive measure than any-frame dicodon freq



Dicodon Preference Model

• The preference value P(X) of a dicodon X is defined as

P(X) = log FC(X)/FN(X)where

FC(X) is freq of X in coding regionsFN(X) is freq of X in non-coding

regions


Dicodon Preference Model’s Properties

• P(X) = 0 if X has same freq in coding and non-coding regions

• P(X) > 0 if X has higher freq in coding than in non-coding region; the larger the diff, the more positive the score is

• P(X) < 0 if X has higher freq in non-coding than in coding region; the larger the diff, the more negative the score is


Dicodon Preference Model Example

• Suppose AAA ATT, AAA GAC, AAA TAG have the following freq:

FC(AAA ATT) = 1.4%FN(AAA ATT) = 5.2%

FC(AAA GAC) = 1.9%FN(AAA GAC) = 4.8%

FC(AAA TAG) = 0.0%FN(AAA TAG) = 6.3%

• ThenP(AAA ATT) = –0.57P(AAA GAC) = –0.40

P(AAA TAG) = –,treating STOP codons

differently

A region consisting of only these dicodons is probably a non-coding region


Frame-Insensitive Coding Region Preference

Model• A frame-insensitive coding preference

Sis(R) of a region R can be defined as

Sis(R) = X is a dicodon in R P(X)

• R is predicted as coding region if Sis(R) > 0

• NB. This model is not commonly used


In-Frame Dicodon Preference Model

• The in-frame + i preference value Pi(X) of a dicodon X is defined as

Pi(X) = log FCi(X)/FN(X)

• where FCi(X) is freq of X in coding regions

at in-frame + i positionsFN(X) is freq of X in non-coding

regions ATG TGC CGC GCT P0

P1

P2


In-Frame Coding Region Preference

Model• The in-frame + i preference Si(R) of a region R

can be defined as

Si(R) = X is a dicodon at in-frame + i position in R Pi(X)

• R is predicted as coding if i=0,1,2 Si(R)/|R| > 0

• NB. This coding preference model is commonly used

• Calculate all ORFs of a DNA segment• For each ORF

– Slide thru ORF w/ increment of 10bp– Calculate in-frame coding region preference

score, in same frame as ORF, within window of 60bp

– Assign score to center of window

• E.g., forward strand in a particular frame...

preference scores

0

+5

-5

Coding Region Prediction: An Example Procedure


Image credit: Xu

• Making the call: coding or non-coding and where the boundaries are

Need training set with known coding and non-coding regions to select threshold that includes as many known coding regions as possible, and at the same time excludes as many known non-coding regions as possible

coding region? where to draw the

boundaries?

where to draw the line?

Problem with Coding Region Boundaries


Image credit: Xu

• Knowing boundaries of coding regions helps identify them more accurately

• Possible boundaries of an exon

• Splice junctions:– Donor site: coding region | GT– Acceptor site: CAG | TAG | coding region

• Translation start– in-frame ATG

{ translation start, acceptor site }

{ translation stop, donor site }

Types of Coding Region Boundaries


Image credit: Xu


• Splice junction sites and translation starts have certain distribution profiles

• For example, ...

Signals for Coding Region Boundaries

• If we align all known acceptor sites (with their splice junction site aligned), we have the following nucleotide distribution

• Acceptor site: CAG | TAG | coding region

Acceptor Site (Human Genome)


Image credit: Xu

• If we align all known donor sites (with their splice junction site aligned), we have the following nucleotide distribution

• Donor site: coding region | GT

Donor Site (Human Genome)

Image credit: Xu



• For a weight matrix, information content of each column is calculated as

– X{A,C,G,T} F(X)*log (F(X)/0.25)

When a column has evenly distributed nucleotides, its information content is lowest

Only need to look at positions having high information content

What Positions Have “High” Information Content?

• Information contentcolumn –3 = – .34*log (.34/.25) – .363*log

(.363/.25) – .183* log (.183/.25) – .114* log (.114/.25) = 0.04

column –1 = – .092*log (.92/.25) – .03*log (.033/.25) – .803* log (.803/.25) – .073* log (.73/.25) = 0.30

Image credit: Xu

Information Content Around Donor Sites in Human

Genome


• Weight matrix model – build a weight matrix for donor, acceptor,

translation start site, respectively– use positions of high information content

Weight Matrix Model for Splice Sites

Image credit: Xu



• Add up freq of corr letter in corr positions:

• Make prediction on splice site based on some threshold

AAGGTAAGT: .34 + .60 + .80 +1.0 + 1.0 + .52 + .71 + .81 + .46 = 6.24

TGTGTCTCA: .11 + .12 + .03 +1.0 + 1.0 + .02 + .07 + .05 + .16 = 2.56

Image credit: Xu

Splice Site Prediction: A Procedure


Other Factors Considered by GRAIL

• G+C composition affects dicodon distributions

• Length of exons follows certain distribution• Other signals associated with coding

regions– periodicity – structure information – .....

• Pseudo genes• ........

Info Fusion by ANN in GRAIL

Image credit: Xu



Remaining Challenges in GRAIL

• Initial exon • Final exon• Indels & frame shifts

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Any Question?

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Indel & Frame-Shift in Coding Regions

• Problem definition• Indel & frameshift identification• Indel correction• An iterative strategy

Some slides here are “borrowed” from Ying Xu


• Indel = insertion or deletion in coding region

• Indels are usually caused by seq errorsATG GAT CCA CAT …..

ATG GAT CA CAT …..

ATG GAT CTCA CAT …..

Indels in Coding Regions



Effects of Indels on Exon Prediction

• Indels may cause shifts in reading frames & affect prediction algos for coding regions

prefscores

exon

indel

Image credit: Xu

• Preferred reading frame is reading frame w/ highest coding score

• Diff DNA segments may have diff preferred reading frames

Segment a coding sequence into regions w/ consistent preferred reading frames corr well w/ indel positions

Indel identification problem can be solved as a sequence segmentation problem!

Key Idea for Detecting Frame-Shift


Image credit: Xu


Frame-Shift Detection by Sequence Segmentation

• Partition seq into segs so that– Chosen frames of adjacent segs are diff– Each segment has >30 bps to avoid small

fluctuations– Sum of coding scores in the chosen frames

over all segments is maximized

• This combinatorial optimization problem can be solved in 6 steps...


Frame-Shift Detection: Step 1

• Given DNA sequence a1 … an

• Define key quantitiesC(i, r, 1) = max score on a1 … ai,

w/ the last segment in frame r C(i, r, 0) = C(i, r, 1) except that

the last seg may have <30 bps



• Determine relationships among the quantities and the optimization problem, viz.

maxr{0, 1, 2}C(i, r, 1) is optimal solution

• Can calculate C(i, r, 0) & C(i, r, 1) from C(i–k, r, 0) & C(i – k, r, 1) for some k > 0


Frame-Shift Detection: Step 2, C(i,r,0)

• To calculate C(i,r,0), there are 3 possible cases for each position i:– Case 1: no indel occurred at position i

– Case 2: ai is an inserted base

– Case 3: a base has been deleted in front of ai

C(i, r, 0) = max { Case 1, Case 2, Case 3 }


• No indel occurs at position i. Then

C(i,r,0) = C(i–1,(2+r) mod 3,0) + P(1+r) mod 3 (ai–5…ai)

a1 a2 …… ai-5 ai-4 ai-3 ai-2 ai-1 ai

di-codon preference

Frame-Shift Detection: Step 2, Case 1


a1 a2 …… ai-6 ai-5 ai-4 ai-3 ai-2 ai-1 ai

di-codon preference


• ai-1 is an inserted base. Then

C(i,r,0) = C(i–2, (r+2) mod 3, 1) + P(1+r) mod 3 (ai–6...a i–2ai)


a1 a2 …… ai-5 ai-4 ai-3 ai-2 ai-1 ai

add a neutral base “C”


• A base has been deleted in front of ai. Then

C(i, r, 0) = C(i–1, (r+1) mod 3, 1) + Pr (ai–5… ai–1C) +

P(1+r) mod 3 (ai–4… ai–1Cai)


C(i, r, 1) = C(i–30, r, 0) +

i–30 < j i–5 P(j+r) mod 3 (aj…aj+5)

a1 a2 …… ai-30 ai-30+1 …… ai

summed di-codon preference

coding score in frame r

Frame-Shift Detection: Step 2, C(i,r,1)

• To calculate C(i,r,1),

Exercise:This formula is not quite right. Fix it.


Frame-Shift Detection: Step 2, Initiation

• Initial conditions,C (k, r, 0) = –, k < 6C (6, r, 0) = P(1+r) mod 3 (a1 … a6)

C(i, r, 1) = –, i < 30

• This is a dynamic programming (DP) algorithm; the equations are DP recurrences


• Calculation of maxr{0, 1, 2}C(i, r, 1) gives an optimal segmentation of a DNA sequence

• Tracing back the transition points---viz. case 2 & case 3---gives the segmentation results

frame 0 frame 1 frame 2


Image credit: Xu

Frame-Shift Detection: Step 4• Determine of coding regions

– For given H1 and H2 (e.g., = 0.25 and 0.75), partition a DNA seq into segs so that each seg has >30 bases & coding values of each seg are consistently closer to one of H1 or H2 than the other

H1

H2

segmentation result


Image credit: Xu


• Overlay “preferred reading-frame segs” & “coding segs” gives coding region predictions regions w/ indels


Image credit: Xu


• If an “insertion” is detected, delete the base at the transition point

• If a “deletion” is detected, add a neutral base “C” at transition point


• We still need to correct the identified indels...

actual indels

predicted indels

What Happens When Indels Are Close Together?

• Our procedure works well when indels are not too close together (i.e., >30 bases apart)

• When indels are too close together, they will be missed...


• Employ an iterative process, vizFind one set of indels and correct them & then iterate until no more indels can be found

actual indels

predicted indels

predicted indelsin iteration 2

Handling Indels That Are Close Together


Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Any Question?

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Modeling & Recognition of

Histone Promoters

Some slides here are “borrowed” from Rajesh Chowdhary

• Play essential role in chromosomal processes– gene transcription,

regulation, – chromosome

condensation, recombination & replication


Histone

• Basic proteins of eukaryotic cell nucleus

• Form a major part of chromosomal proteins

• Help in packaging DNA in the chromatin complex

• Five types, namely H1, H2A, H2B, H3 and H4

• Highly conserved across species– H1 least conserved, H3

& H4 most conserved

Histone Transcription• TFs bound in core,

proximal, distal promoter & enhancer regions

• TFIID binds to TATA box & identifies TSS with help of TAFs & TBP

• RNA Pol-II supplemented by GTFs (A,B,D,E,F,H) recruited to core promoter to form Pre-initiation complex

• Transcription initiated– Basal/Activated, depending

on space & time


Histone Promoter ModelingWerner 1999

• Three promoter types: Core, proximal and distal• Characterised by the presence of specific TFBSs

– CAAT box, TATA Box, Inr, & DPE– Order and mutual distance of TFBS modules is specific

& determine function


Histone H1t Gene RegulationGrimes et al. 2003

• One gene can express in diff ways in diff cells

• Same binding site can have diff functions in diff cells



Why Model Histone promoters

• To understand histone’s regulatory mechanism – To characterise regulatory features from

known promoters– To identify promoter from uncharacterised

genomic sequence (promoter recognition)– To find other genes with similar regulatory

behaviour and gene-products – To define potential gene regulatory networks


Difficulties of Histone Promoter Modeling

• Not a plain sequence alignment problem• Not all features are common among

different groups• Not only TFBSs’ presence, but their

location, order, mutual distance and orientation are critical to promoter function

• Not all TFs & TFBSs have been characterized yet


Tools for Promoter Modeling

• Genomic signals in promoter v/s non-promoter– Core promoter (TATA

Box, Inr, DPE) and/or few TFBS outside core promoter

– Entire promoter (core, proximal & distal) with whole ensemble of TFBS

• Genomic content in promoter v/s non-promoter – CpG islands, GC content

• 2D-3D DNA structural features

• Model with a scoring system based on training data (good data not always available)– Input seq scanned for

desired patterns & those whose scores above certain threshold are reported

Promoter Recognition Programs

• Programs have different objectives

• Use various combinations of genomic signals and content

• Typically analyse 5’ region [-1000,+500]

• Due to low accuracy, programs developed for sub-classes of promoters


Image credit: Rajesh


Steps for Building Histone Promoter Recognizer

• Exercise: What do you think these steps are?


MEME

• MEME is a powerful and good method for finding motifs from biological sequences

• T. L. Bailey & C. Elkan, "Fitting a mixture model by expectation maximization to discover motifs in

biopolymers", ISMB, 2:28--36, 1994

H2A

Motifs Discovered by MEME in Histone Gene 5’ Region [-

1000,+500]



H2B

Motifs Discovered by MEME in Histone Gene 5’ Region [-

1000,+500]



Are These Really Motifs of H2A and H2B Promoters?


H2B

H2A

• One could use the motifs discovered by MEME to detect H2A & H2B promoters

• But….it is strange that the motifs for H2A and H2B are generally the same, but in opposite orientation

• Exercise: Suggest a possible explanation Image credit: Rajesh

The Real Common Promoter Region of H2A & H2B is at [-

250,-1]!


H2B

H2A

• MEME was overwhelmed by coding region & did not find the right motifs! Image credit: Rajesh

Motifs Discovered by MEME in Histone Promoter 5’ Region [-

250,-1]• Discovered 9 motifs

among all 127 histone promoters

• All 9 motifs are experimentally proven TFBSs (TRANSFAC)



Deriving Histone Promoter Models

• Divide H1 seqs into 5 subgroups

• Aligned seqs within each subgroup

• Consensus alignment matches biologically known H1 subgroup models

Can apply same approach to find promoter models for H2A, H2B, H3, H4...



Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Any Question?

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Knowledge Discovery Basics

• Knowledge discovery in brief• K-Nearest Neighbour• Support Vector Machines• Bayesian Approach• Hidden Markov Models• Artificial Neural Networks Some slides are from a tutorial jointly taught with Jinyan Li

Jonathan’s rules : Blue or CircleJessica’s rules : All the rest

Whose block is this?

Jonathan’s blocks

Jessica’s blocks

What is Knowledge Discovery?


Image credit: Tan

Question: Can you explain how?

What is Knowledge Discovery?


Image credit: Tan


Some classifiers/learning methods

Steps of Knowledge Discovery

• Training data gathering• Feature generation

– k-grams, colour, texture, domain know-how, ...

• Feature selection– Entropy, 2, CFS, t-test, domain know-how...

• Feature integration– SVM, ANN, PCL, CART, C4.5, kNN, ...


Some Knowledge Discovery Methods

• K-Nearest Neighbour• Support Vector Machines• Bayesian Approach• Hidden Markov Models• Artificial Neural Networks


• A common “distance” measure betw samples x and y is

where f ranges over features of the samples

How kNN Works

• Given a new case• Find k “nearest”

neighbours, i.e., k most similar points in the training data set

• Assign new case to the same class to which most of these neighbours belong

Neighborhood

5 of class

3 of class

=

Illustration of kNN (k=8)


Image credit: Zaki

Enzyme inducers

Peroxisome proliferators


Prediction of Compound Signature Based on Gene

Expression Profiles• Hamadeh et al, Toxicological

Sciences 67:232-240, 2002

• Store gene expression profiles corr to biological responses to exposures to known compounds whose toxicological and pathological endpoints are well characterized

• use kNN to infer effects of unknown compound based on gene expr profiles induced by it


(a) Linear separation not possible w/o errors(b) Better separation by nonlinear surfaces in input space(c ) Nonlinear surface corr to linear surface in feature

space. Map from input to feature space by “kernel” function

“Linear learning machine” + kernel function as classifier

Basic Idea of SVMImage credit: Zien

• Hyperplane separating the x’s and o’s points is given by (W•X) + b = 0, with (W•X) = jW[j]*X[j]

Decision function is llm(X) = sign((W•X) + b))

Linear Learning Machines



• Solution is a linear combination of training points Xk with labels Yk

W[j] = kk*Yk*Xk[j],

with k > 0, and Yk = ±1

llm(X) = sign(kk*Yk* (Xk•X) + b)

Linear Learning Machines

“data” appears only in dot product!

• llm(X) = sign(kk*Yk* (Xk•X) + b)

• svm(X) = sign(kk*Yk* (Xk• X) + b)

svm(X) = sign(kk*Yk* K(Xk,X) + b)

where K(Xk,X) = (Xk• X)

Kernel Function



Kernel Function

• svm(X) = sign(kk*Yk* K(Xk,X) + b) K(A,B) can be computed w/o computing

• In fact replace it w/ lots of more

“powerful” kernels besides (A • B). E.g.,– K(A,B) = (A • B)d

– K(A,B) = exp(– || A B||2 / (2*)), ...


How SVM Works

• svm(X) = sign(kk*Yk* K(Xk,X) + b)

• To find k is a quadratic programming problem

max: kk – 0.5 * k h k*h Yk*Yh*K(Xk,Xh)

subject to: kk*Yk=0

and for all k , C k 0

• To find b, estimate by averagingYh – kk*Yk* K(Xh,Xk)

for all h 0

Prediction of Gene Function From Gene Expression Data

Using SVM• Brown et al., PNAS

91:262-267, 2000

• Use SVM to identify sets of genes w/ a c’mon function based on their expression profiles

• Use SVM to predict functional roles of uncharacterized yeast ORFs based on their expression profiles



Bayes Theorem

• P(h) = prior prob that hypothesis h holds• P(d|h) = prob of observing data d given h holds• P(h|d) = posterior prob that h holds given

observed data d


• Let H be all possible classes. Given a test instance w/ feature vector {f1 = v1, …, fn = vn}, the most probable classification is given by

• Using Bayes Theorem, rewrites to

• Since denominator is indep of hj, simplifies to

Bayesian Approach


Naïve Bayes

• But estimating P(f1=v1, …, fn=vn|hj) accurately may not be feasible unless training data set is sufficiently large

• “Solved” by assuming f1, …, fn are indep

• Then

• where P(hj) and P(fi=vi|hj) can often be estimated reliably from typical training data set

Bayesian Design of Screens for Macromolecular

Crystallization• Hennessy et al., Acta Cryst

D56:817-827, 2000

• Xtallization of proteins requires search of expt settings to find right conditions for diffraction-quality xtals

• BMCD is a db of known xtallization conditions

• Use Bayes to determine prob of success of a set of expt conditions based on BMCD


How HMM Works

• HMM is a stochastic generative model for sequences

• Defined by– finite set of states S– finite alphabet A– transition prob matrix T– emission prob matrix E

• Move from state to state according to T while emitting symbols according to E

sk

s1

…

s2

a1

a2



• In nth order HMM, T & E depend on all n previous states

• E.g., for 1st order HMM, given emissions X = x1, x2, …, & states S = s1, s2, …, the prob of this seq is

• If seq of emissions X is given, use Viterbi algo to get seq of states S such that

S = argmaxS Prob(X, S)

• If emissions unknown, use Baum-Welch algo

How HMM Works


• Game:– You bet $1– You roll – Casino rolls – Highest number wins

$2

• Question: Suppose we played 2 games, and the sequence of rolls was 1, 6, 2, 6. Were we likely to be cheated?

Example: Dishonest Casino

• Casino has two dices:– Fair dice

• P(i) = 1/6, i = 1..6

– Loaded dice• P(i) = 1/10, i = 1..5• P(i) = 1/2, i = 6

• Casino switches betw fair & loaded die with prob 1/2. Initially, dice is always fair

“Visualization” of Dishonest Casino



1, 6, 2, 6? We were probably cheated...

Protein Families Modelling By HMM

• Baldi et al., PNAS 91:1059-1063, 1994

• HMM is used to model families of biological sequences, such as kinases, globins, & immunoglobulins

• Bateman et al., NAR 32:D138-D141, 2004

• HMM is used to model 6190 families of protein domains in Pfam


What are ANNs?

• ANNs are highly connected networks of “neural computing elements” that have ability to respond to input stimuli and learn to adapt to the environment...


Computing Element

• Behaves as a monotone function y = f(net), where net is cummulative input stimuli to the neuron

• net is usually defined as weighted sum of inputs

• f is usually a sigmoid


How ANN Works

• Computing elements are connected into layers in a network

• The network is used for classification as follows:– Inputs xi are fed into

input layer– each computing element

produces its corr output– which are fed as inputs

to next layer, and so on– until outputs are

produced at output layer

• What makes ANN works is how the weights on the links are learned

• Usually achieved using “back propagation”


vij

wjzj

Back Propagation• vji = weight on link

betw xi and jth computing element in 1st layer

• wj be weight of link betw jth computing element in 1st layer and computing element in last layer

• zj = output of jth computing element in 1st layer

• Then


Back Propagation

• For given sample, y may differ from target output t by amt

• Need to propagate this error backwards by adjusting weights in proportion to the error gradient

• For math convenience, define the squared error as

To find an expression for weight adjustment, we differentiate E wrt vij and wj to obtain error gradients for these weightsvij

wjzj



Applying chain rule a few times and recalling definitions of y, zj, E, and f, we derive...

Back Propagation

vij

wjzj


T-Cell Epitopes Prediction By ANN

• Honeyman et al., Nature Biotechnology 16:966-969, 1998

• Use ANN to predict candidate T-cell epitopes


Image credit: Brusic

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Translation Initiation Site Recognition

An introduction to the World’s simplest TIS recognition system

Translation Initiation Site



299 HSU27655.1 CAT U27655 Homo sapiensCGTGTGTGCAGCAGCCTGCAGCTGCCCCAAGCCATGGCTGAACACTGACTCCCAGCTGTG 80CCCAGGGCTTCAAAGACTTCTCAGCTTCGAGCATGGCTTTTGGCTGTCAGGGCAGCTGTA 160GGAGGCAGATGAGAAGAGGGAGATGGCCTTGGAGGAAGGGAAGGGGCCTGGTGCCGAGGA 240CCTCTCCTGGCCAGGAGCTTCCTCCAGGACAAGACCTTCCACCCAACAAGGACTCCCCT............................................................ 80................................iEEEEEEEEEEEEEEEEEEEEEEEEEEE 160EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE 240EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

A Sample cDNA

• What makes the second ATG the TIS?


Approach

• Training data gathering• Signal generation

– k-grams, distance, domain know-how, ...

• Signal selection– Entropy, 2, CFS, t-test, domain know-how...

• Signal integration– SVM, ANN, PCL, CART, C4.5, kNN, ...


Training & Testing Data

• Vertebrate dataset of Pedersen & Nielsen [ISMB’97]

• 3312 sequences• 13503 ATG sites• 3312 (24.5%) are TIS• 10191 (75.5%) are non-TIS• Use for 3-fold x-validation expts


Signal Generation

• K-grams (ie., k consecutive letters)– K = 1, 2, 3, 4, 5, …– Window size vs. fixed position– Up-stream, downstream vs. any where in window– In-frame vs. any frame

0

0.5

1

1.5

2

2.5

3

A C G T

seq1

seq2

seq3


299 HSU27655.1 CAT U27655 Homo sapiensCGTGTGTGCAGCAGCCTGCAGCTGCCCCAAGCCATGGCTGAACACTGACTCCCAGCTGTG 80CCCAGGGCTTCAAAGACTTCTCAGCTTCGAGCATGGCTTTTGGCTGTCAGGGCAGCTGTA 160GGAGGCAGATGAGAAGAGGGAGATGGCCTTGGAGGAAGGGAAGGGGCCTGGTGCCGAGGA 240CCTCTCCTGGCCAGGAGCTTCCTCCAGGACAAGACCTTCCACCCAACAAGGACTCCCCT

Signal Generation: An Example

• Window = 100 bases• In-frame, downstream

– GCT = 1, TTT = 1, ATG = 1…

• Any-frame, downstream– GCT = 3, TTT = 2, ATG = 2…

• In-frame, upstream– GCT = 2, TTT = 0, ATG = 0, ...


Too Many Signals

• For each value of k, there are4k * 3 * 2 k-grams

• If we use k = 1, 2, 3, 4, 5, we have4 + 24 + 96 + 384 + 1536 + 6144

= 8188features!

• This is too many for most machine learning algorithms

Signal Selection (Basic Idea)

• Choose a signal w/ low intra-class dist

• Choose a signal w/ high inter-class dist


Image credit: Slonim


Signal Selection (eg., t-statistics)


Signal Selection (eg., 2)


Signal Selection (eg., CFS)

• Instead of scoring individual signals, how about scoring a group of signals as a whole?

• CFS– Correlation-based Feature Selection– A good group contains signals that are highly

correlated with the class, and yet uncorrelated with each other


• Position –3• in-frame upstream ATG• in-frame downstream

– TAA, TAG, TGA, – CTG, GAC, GAG, and GCC

Kozak consensusLeaky scanning

Stop codon

Codon bias?

Sample k-grams Selected by CFS


Signal Integration

• kNN– Given a test sample, find the k training

samples that are most similar to it. Let the majority class win.

• SVM– Given a group of training samples from two

classes, determine a separating plane that maximises the margin of error.

• Naïve Bayes, ANN, C4.5, ...


TP/(TP + FN) TN/(TN + FP) TP/(TP + FP) Accuracy

Naïve Bayes 84.3% 86.1% 66.3% 85.7%

SVM 73.9% 93.2% 77.9% 88.5%

Neural Network 77.6% 93.2% 78.8% 89.4%

Decision Tree 74.0% 94.4% 81.1% 89.4%

Results (3-fold x-validation)



NB 84.3% 86.1% 66.3% 85.7%

SVM 73.9% 93.2% 77.9% 88.5%

NB+Scanning 87.3% 96.1% 87.9% 93.9%

SVM+Scanning 88.5% 96.3% 88.6% 94.4%

Improvement by Scanning

• Apply Naïve Bayes or SVM left-to-right until first ATG predicted as positive. That’s the TIS.

• Naïve Bayes & SVM models were trained using TIS vs. Up-stream ATG



NB 84.3% 86.1% 66.3% 85.7%

Decision Tree 74.0% 94.4% 81.1% 89.4%

NB+NN+Tree 77.6% 94.5% 82.1% 90.4%

SVM+Scanning 88.5% 96.3% 88.6% 94.4%*

Pedersen&Nielsen 78% 87% - 85%

Zien 69.9% 94.1% - 88.1%

Hatzigeorgiou - - - 94%*

* result not directly comparable

Performance Comparisons

F

L

I

MV

S

P

T

A

Y

H

Q

N

K

D

E

C

WR

G

A

T

E

L

R

S

stop

How about using k-grams from the translation?

mRNAprotein



Amino Acid K-grams Discovered (by entropy)


Independent Validation Sets

• A. Hatzigeorgiou:– 480 fully sequenced human cDNAs– 188 left after eliminating sequences similar to

training set (Pedersen & Nielsen’s)– 3.42% of ATGs are TIS

• Our own:– well characterized human gene sequences

from chromosome X (565 TIS) and chromosome 21 (180 TIS)

Validation Results (on Hatzigeorgiou’s)

• Using top 100 features selected by entropy and trained on Pedersen & Nielsen’s dataset


ATGpr

Ourmethod

Validation Results (on Chr X & Chr 21)

• Using top 100 features selected by entropy and trained on Pedersen & Nielsen’s


Image credit: Liu

Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Human Polyadenylation Signal Prediction

Some slides are “borrowed” from Huiqing Liu


Cleavage & Polyadenylation of Pre-mRNAs in Mammalian

Cells

PAS in Human Pre-mRNA 3’end Processing Site

• Selection of poly-A site is primarily determined by a hexameric poly-A signal (PAS) of sequence AAUAAA (or a one-base variant) & downstream U-rich or GU-rich elements


PAS Prediction: Step 1

Feature generati on

I ncomi ngsequences

Feature sel ecti on

Feature i ntegrati on

END

BEGI N



discard all the features without cut points.

Entropy Measure for Feature Selection

Fayyad & Irani, IJCAI, 1993

Feature generati on

I ncomi ngsequences

Feature sel ecti on


END

BEGI N




SVM in Weka

Feature generati on

I ncomi ngsequences

Feature sel ecti on


END

BEGI N


Data Set I (From Erpin)

• Training set: 2327 seqs– 1632 “unique” & 695

“strong” poly(A) sites

• All seq trimmed to contain 206 bases, having a false or true PAS in the center

• Positive testing set: – 982 seq w/ annotated

PASes from EMBL

• Negative testing set:– 982 CDS seqs– 982 seqs of 1st intron– 982 randomized UTR

seqs using same 1st order Markov model as human 3’ UTRs

– 982 randomized UTR seqs of same mono nucleotide composition as human 3’ UTRs


Data Set II (mRNA data)

• Positive set: – 312 human mRNA seqs

from RefSeq release 1– Each contains a

“poly(A)-signal” feature tag carrying an “evidence=experimental” label

– 767 human mRNA sequences from RefSeq containing a “poly(A)-site” feature tag carrying an “evidence=experimental” label. Similar sequences have been removed

• Negative set: – Generated by scanning

“AATAAA” at coding region (exclude those near the end of seq)

Experimental Results

• Preliminary test: In order to compare with the performance of Erpin and Polyadq, we also adjust prediction accuracy on 982 true PASes at around 56%.


All the numbers regarding to the performance of Erpin and Polyadq are copied or derived from Legendre & Gautheret 2003


• Testing results on Erpin using validation sets


It is clear that both upstream and dowstream are well-characterized by G/U rich segments (consistent w/ reported motifs)


• Top ranked features


Cop

yrig

ht ©

200

4 by

Lim

soon

Won

g

Recognition of Transcription Start

Sites

An introduction to the World’s best TSS

recognition system: A heavy tuning approach


Transcription Start Site


-200 to +50window size

Model selected based on desired sensitivity

Structure of Dragon Promoter Finder

Image credit: Bajic

GC-rich submodel

GC-poor submodel

(C+G) =#C + #GWindow Size


Each model has two submodels based on GC

content

Image credit: Bajic

Pentamer at ith

position in input

jth pentamer atith position in training window

Frequency of jthpentamer at ith positionin training window

Window size

Promoter, Exon, Intron Sensors

• These sensors are positional weight matrices of k-grams, k = 5 (aka pentamers)

• They are calculated as s below using promoter, exon, intron data respectively


Tuning parameters

tanh(x) =ex e-x

ex e-x

sIE

sI

sEtanh(net)

Simple feedforward ANN trained by the Bayesian regularisation method

wi

net = si * wi

Tunedthreshold

Data Preprocessing & ANN


without C+G submodels

with C+G submodels

Accuracy Comparisons


Image credit: Bajic


Acknowledgements

I “borrowed” a lot of materials in this lecture from

• Xu Ying, Univ of Georgia• Mark Craven, Univ of Wisconsin• Ken Sung, NUS• Rajesh Chowdhary, I2R• Jinyan Li, I2R• Huiqing Liu, I2R


Primary References• Y. Xu et al. “GRAIL: A Multi-agent neural network system

for gene identification”, Proc. IEEE, 84:1544--1552, 1996• R. Staden & A. McLachlan, “Codon preference and its use

in identifying protein coding regions in long DNA sequences”, NAR, 10:141--156, 1982

• Y. Xu, et al., "Correcting Sequencing Errors in DNA Coding Regions Using Dynamic Programming", Bioinformatics, 11:117--124, 1995

• Y. Xu, et al., "An Iterative Algorithm for Correcting DNA Sequencing Errors in Coding Regions", JCB, 3:333--344, 1996

• R. Chowdhary et al., “Modeling 5' regions of histone genes using Bayesian Networks”, APBC 2005, accepted


Primary References• H. Liu et al., "Data Mining Tools for Biological Sequences",

JBCB, 1:139--168, 2003• H. Liu et al., "An in-silico method for prediction of

polyadenylation signals in human sequences", GIW, 14:84--93, 2003

• V. B. Bajic et al., "Dragon Gene Start Finder: An advanced system for finding approximate locations of the start of gene transcriptional units", Genome Research, 13:1923--1929, 2003


Other Useful Readings• L. Wong. The Practical Bioinformatician. World Scientific,

2004• T. Jiang et al. Current Topics in Computational Molecular

Biology. MIT Press, 2002• R. V. Davuluri et al., "Computational identification of

promoters and first exons in the human genome", Nat. Genet., 29:412--417, 2001

• J. E. Tabaska et al., "Identifying the 3'-terminal exon in human DNA", Bioinformatics, 17:602--607, 2001

• J. E. Tabaska et al., "Detection of polyadenylation signals in human DNA sequences", Gene, 23:77--86, 1999

• A. G. Pedersen & H. Nielsen, “Neural network prediction of translation initiation sites in eukaryotes”, ISMB, 5:226--233, 1997


Other Useful Readings• C. Burge & S. Karlin. “Prediction of Complete Gene

Structures in Human Genomic DNA”, JMB, 268:78--94, 1997• V. Solovyev et al. "Predicting internal exons by

oligonucleotide composition and discriminant analysis of spliceable open reading frames", NAR, 22:5156--5163, 1994

• V. Solovyev & A. Salamov. “The Gene-Finder computer tools for analysis of human and model organisms genome sequences", ISMB, 5:294--302, 1997

• T. A. Down & T. J. P. Hubbard. “Computational Detection and Location of Transcription Start Sites in Mammalian Genomic DNA”, Genome Research, 12:458--461, 2002

• T. L. Bailey & C. Elkan, "Fitting a mixture model by expectation maximization to discover motifs in biopolymers", ISMB, 2:28--36, 1994


Other Useful Readings• A. Zien et al., “Engineering support vector machine kernels that

recognize translation initiation sites”, Bioinformatics, 16:799--807, 2000

• A. G. Hatzigeorgiou, “Translation initiation start prediction in human cDNAs with high accuracy”, Bioinformatics, 18:343--350, 2002

• V.B.Bajic et al., “Computer model for recognition of functional transcription start sites in RNA polymerase II promoters of vertebrates”, J. Mol. Graph. & Mod., 21:323--332, 2003

• J.W.Fickett & A.G.Hatzigeorgiou, “Eukaryotic promoter recognition”, Genome Research, 7:861--878, 1997

• A.G.Pedersen et al., “The biology of eukaryotic promoter prediction---a review”, Computer & Chemistry, 23:191--207, 1999

• M.Scherf et al., “Highly specific localisation of promoter regions in large genome sequences by PromoterInspector”, JMB, 297:599--606, 2000


Other Useful Readings• M. A. Hall, “Correlation-based feature selection machine

learning”, PhD thesis, Univ. of Waikato, New Zealand, 1998• U. M. Fayyad, K. B. Irani, “Multi-interval discretization of

continuous-valued attributes”, IJCAI, 13:1022-1027, 1993• H. Liu, R. Sentiono, “Chi2: Feature selection and discretization

of numeric attributes”, IEEE Intl. Conf. Tools with Artificial Intelligence, 7:338--391, 1995

• C. P. Joshi et al., “Context sequences of translation initiation codon in plants”, PMB, 35:993--1001, 1997

• D. J. States, W. Gish, “Combined use of sequence similarity and codon bias for coding region identification”, JCB, 1:39--50, 1994

• G. D. Stormo et al., “Use of Perceptron algorithm to distinguish translational initiation sites in E. coli”, NAR, 10:2997--3011, 1982

• Legendre & Gautheret, “Sequence determinants in human polyadenylation site selection”, BMC Genomics, 4(1):7, 2003

copyright © 2004 by limsoon wong gene finding & gene feature recognition by computational...

Documents

amino acidslength

amino acids folds

amino acids average

limsoon wongorientation

limsoon wongnucleotideconsists

form hydrogen bonds

hydrogenbonds copyright

model of dna