Cédric Notredame (22/04/2015) Finding Genes In a Genome Cédric Notredame.

Cédric Notredame (18/04/23)

Finding Genes

In a Genome

Cédric Notredame


Naked Genome


All Dressed Up!




Naked Genomes are Useless

Useful Genome

Accurate Annotation

-Experimental Methods

-Computational Methods

-ESTs, THS, DNA Chips…

-Homology, Ab-Initio


ANNOTATION

-Where are the genes ?

-What do they do: Biochemistry ?

-When do they do it: Regulation ?

-Who do they do it for: Metabolic ?


Outline

Naked Genome => Fully Dressed Sequence

1. Cleaning the genome

2. Similarity methods

3. Experimental Methods

4. Ab-initio MethodsEukaryotes

Prokaryotes

5-How Good Are The Methods ??


Outline

Eukaryotes

Prokaryotes



Gene Fishingin

Prokaryotic Genomes


What is a Prokaryotic Gene ?

GenePromoter RBS

Protein

ORF

mRNASTOPATG

Terminator


What is a Prokaryotic Gene:Operon


2-Homology Based Methods

1-Ab-initio:-ORFing-Codon Bias

Promoter RBS

mRNASTOP

Terminator

3-Regulatory Sequence Detection-Non Coding-Short Genes


Prokaryotic Genomes

-High Gene Density:Haemophilus Influenza: 85%

-No Introns

-Operons

In a prokaryotic Genome, any ORF longer than 300 nt

Can SAFELY be considered to be a gene


Prokaryotic Genomes

Clean-up

ORFing

Homology Search

Gene Prediction

Promoter Detection


CleaningYourDNA

Sequence


Cleaning a DNA Sequence

Is My Sequence Contaminated ?

-Cloning may lead tothe inclusion of VectorSequences.

-These sequences must beremoved


Paste in yournew sequence


Crop

Our sequence displays two vectorcontaminations


Contamination Matters

Contaminations Look Like Horizontal Transfers

BUT Genuine Genome may Contain Similarity tothe Cloning vector (Antibiotics Resistance)

-Wrong Phylogeny

-Error Propagation in Secondary Databases

-Eukaryote Genomes can also be cleaned this way


ORFingProkaryotic Genomes


Prokaryotic Genomes: ORFing

Where are the ORFs In my Sequence ?



ATG (Start) Codons

STOP Codons




Prokaryotic Genomes: GORF

www.ncbi.nih.gov/gorf/gorf.html





TO COG

TO BLAST




GORF: Can You Trust it ???

Random ORF Random 3rd Position

Real ORF Biased 3rd Position


GORF: Can You Trust it ???


Prokaryotic Genomes: GORFing cDNAs

BUT…

-Will NOT detect SHORT genes

-Will NOT detect Non Coding Genes

Works with Bacterial Genomes

Good enough for ~85% proteome

Works with Eukaryotic cDNA


Ab-InitioGene Predictions

InProkaryotic Genomes


Predicting Genes

What are the sequences in my genome that LOOK LIKE Genes


Using The Codon Biases


Using The Codon Biases

Coding RegionsDo NOT look LikeRandom DNA:

-Codon Bias


Real Genes Use Mostly the Optimal Codons


Predicting Genes

ALL the characteristics of a Gene can be Built into a model

Hidden Markov Model


Hidden Markov Model

-Each Nucleotide has a STATE: Coding/Non Coding …

-This STATE is HIDDEN

-The HMM tries to UNCOVER the STATE of each Nucleotide.


Hidden Markov Model

Occasionally Dishonest CAsino …

-This STATE is HIDDEN in the data

Observation: 122234455666125654151661661515566616166661

State : FFFFFFFFLLLLFFFFFFLLLLLLLLLLLLLLLLLLLLLLLLLLL


GeneMark


Simplified HMM for Coding Regions

S

GGG 0.02GGGA 0.00GGGT 0.6GGGC 0.38G

TGG 1.00W

64 Codons

GGG 0.02GGGA 0.00GGGT 0.6GGGC 0.38G

TGG 1.00W

64 Codons

E


Emission Proba

Transition Proba



Proba of seq (GGG-TGG Given Model)

=Proba(GGG)*Proba(GGG-

>TGG)*Proba(TGG)

HMM order 5: 6th Nucleotide depends on the 5 previous

Takes into account Codon Bias AND dipeptide Comp


Cédric Notredame (18/04/23)Translate Predicted Genes into Proteins Text Output

http://opal.biology.gatech.edu/GeneMark/



Non Standard FASTA


GLIMMER: An alternative to GeneMark


Main Problems


GeneMark and HMM predictions

Works Very Well

Good enough for ~99% proteome

BUT…

-Will NOT detect Some SHORT genes

-Will NOT detect Non Coding Genes


Which Program ???

The established programs ALL work well

No point in fighting if your users have their mind set on a brand…


If Your Gene is NON-Coding…

The only existing model for NON-Coding genes are those for tRNA


Homology BasedGene Prediction

InProkaryotic Genomes


BLASTx

What are the portion of my GenomeThat Look like a Known Gene/Protein?


blastx

protein

nucleotide

proteinVS

Non Coding, but works only for higly similar sequences ( >70%)


BlastX and HMM predictions

BUT…

Needs Homology

Depends on the databases

Very Reliable on Prokaryotes

Can Help in Eukaryotes


Finding PromotersIn

Prokaryotic Genomes


Promoter Hunting

Are There known promoters in my Sequence ?

Ideal for

-Finding Small Proteins

-Finding Non Coding Genes






prodoric.tu-bs.de/


prodoric.tu-bs.de/



rsat.ulb.ac.be/rsat/RSA_home.cgi



Fishing GenesIn

Eukaryotic Genomes



2-Homology Based Methods

1-Transcript Based Methods 3-Ab-initio:-HMMs

4-Regulatory Sequence Detection

Promoter

mRNA (form2)

exonexon exon exon exonexon

mRNA (form2)


Eukaryote Genomes

Clean-up

Transcripts

Prediction

Homology

Promoter Detection


Know your

Opponent …


Exons are longer in Vertebrates


Introns are longer in Vertebrates-100 bp in Fungi-1000 bp in Vertebrates


Genes contain more Introns in Mammals


Cleaning

Eukaryotic Genomes


RepeatsRepeatsTransposable elements, simple repeatsTransposable elements, simple repeats

RepeatMaskerRepeatMaskerSmith and Waterman Clean-up.Smith and Waterman Clean-up.

Avoiding Repeats

PlusPlus -Remove lots of noise.-Remove lots of noise.

MinusMinus -Changes Sequence -Changes Sequence Statistics.Statistics.


Homology BasedGene Prediction

InEukaryotic Genomes


Homology Based Predictions

What are the portion of my GenomeThat Look like a Known Protein?


Three Tools

GeneWise:GeneWise: Most CommonMost Common

Procrustes:Procrustes: Most SophisticatedMost Sophisticated

BlastX/TBlastXBlastX/TBlastX SimplestSimplest


blastx

protein

Genome

proteinVS

BLASTX


tblastx

protein

Genome

protein

ESTs

VS

TBLASTX: Exon Fishing


genomic sequence

Protein

Procrustes


40% id


www.ebi.ac.uk/Wise2/advanced.html

GeneWise

genomic sequence

Protein


Transcript BasedGene Prediction



Gene indices

Using Established ESTs Collections


AAAAAA...putativemRNA exon 15‘UTR exon 2 3‘UTR

expressed sequence tags(ESTs)

ESTs give us an Insight into this Complexity

1-Cluster the ESTs to reconstitute a gene


EMBL database

Quality clipping

BLASTsearch, clustering

EST Collection

Quality clipping

Assembly,Consensus sequences

Visualization

Gene indices Typical WorkFlow


Gene indices Alignment

consensus


Gene indices Alignment Software

Phrap (Phil Green)

CAP3 (X. Huang)

TIGR assembler

GAP4 (R. Staden)


Gene indicesConsensus sequences

Reduced error rate

Long Consensus

Efficient database search

exon/intron boundaries

Alternative Splicing


UniGene (NCBI)

TIGR Gene Indices

STACK (SANBI)

GeneNest (DKFZ,MPI)

Goal:

One cluster

One Gene

Gene indices

Clustering of EST and mRNA sequences of an organism to reduce redundance in sequence data.


GeneNestgenenest.molgen.mpg.de


TIGR Gene Indices

Alignment scheme

www.tigr.org


UniGene

www.ncbi.nih.nlm.gov/UniGene


UniGene

www.ncbi.nih.nlm.gov/UniGene


Gene indices


Gene indicesApplications

Detection of exon/intron boundaries

Detection of alternative splicing

Detection of Single Nucleotide Polymorphisms

Genome annotation

Analysis of gene expression

Design of DNA-chips/arrays


Mapping of EST consensus sequences on genomic DNA

genomic sequence

exons

consensus sequence( mRNA)


Comparing YourComparing YourGenomeGenome

with Transcriptswith Transcripts

HowHow to to dodo It ? It ?How Long ?How Long ?

BLAST : 36 hoursBLAST : 36 hours

Popular and well Popular and well describeddescribed

HSPs tend to mangle HSPs tend to mangle IntronsIntronsEST_GENOME 80 hoursEST_GENOME 80 hours

Dynamic Program. post processDynamic Program. post process

Slow and sometimes hard to useSlow and sometimes hard to use

BLAT: 0.5 hoursBLAT: 0.5 hours

Next GenerationNext Generation

Look for nearly identical seq.Look for nearly identical seq.

SIM4 pbil.univ-lyon1.fr/sim4.phpSIM4 pbil.univ-lyon1.fr/sim4.php

Similar to BLAT (slower)Similar to BLAT (slower)

Allows Large GapsAllows Large Gaps



Mapping cDNA on genomic DNA

splicenest.molgen.mpg.de


Gene indicesApplications

Detection of exon/intron boundaries

Detection of alternative splicing

Detection of Single Nucleotide Polymorphisms

Genome annotation

Analysis of gene expression

Design of DNA-chips/arrays



genomic sequence

exons

consensus sequence( mRNA)

splice variant



Splice variants of APECED gene

number of sequences genomic sequencealternative variants




Alternative Splicing(additional exon)

1-skipped exon

Splice variants of adenylsuccinate lyase

2-unspliced ?

3-gene prediction errors ?



Alternative Splicing (alternative donor site)


Alternative Splicing(unknown gene Hs16936)


Ab-InitioGene Prediction




Three Categories of Methods

•Rule Based

–Uses explicit set of rules to make decisions

–GeneFinder•Neural Network

–Uses a data set to build rules.

–Grail•HMM

–Finding the state of each Nucleotide (Coding…)

–Genscan


Rule Based - GeneFinder

• CodonBias -> score1

• Splice Site description -> score2

• ORFs -> score3

Proba (Gene)= F(score1, score2, score3..)


• Train Neural Network on Known Genes: Discriminate

• GrailExp : • Measure several coding potentials

• Blast• Coding Potential• …

• Feed all the scores into the Neural Network

Neural Networks - Grail



• Genscan models the genes with a Hidden Markov model, that models coding and non-coding regions.

HMM-Genscan

•Fifth order inhomogeneous HMM

–Fifth order : use 6-tuples (two codons)

–Inhomogeneous: each position is special (0,1,2)


Burge, C. and S. Karlin, Prediction of complete gene structures in human genomic DNA. J Mol Biol, 1997. 268(1): p. 78-94


Your Genomic Sequence

A Collection of Proteins



Evaluating Eukaryote

Gene Prediction


PMID:11042160


Nucleotide Accuracy

TPTP FN

Sn Sp TPTP FP++

TP FN+ TP FP+ TN FP+ TN FN+

= =

TN+TN+TP+TPAC=0.5*( ) -1

((TP+FN)*(TN+FP)*(TP+FP)*(TN+FN))1/2

(TN*TP)+(FN*FP)CC=

sensitivity specificity

approximate correlation

correlation coefficient


Exon Accuracy


•Blastx:

•Good Gene Hunter/Poor Modeler

•GeneWise

•Best Homology Gene Modeling


•GenScan:

•Distracted by Complete Genome

•Use GenomeScan instead


•GenScan:

•Ab-Initio Methods are more Robust


http://www.cs.ubc.ca/~rogic/evaluation.html

PMID: 8786136



•High and Low GC contents can confuse Predictions



Annnnnnd The

Winner is …


http://www.cbs.dtu.dk/services/HMMgene/


Working on a Genome


www.sanger.ac.uk/Software/Artemis/



Wrapping It Up


Predicting Genes

Using Homology: BlastX,

Procrustes

ORFing: GORF

Cleaning Up Data


Predicting Genes

Promoter PredictionPRODORIC

Transcript Based Predictions

GeneNest, UniGene, BLAT

Ab-Initio Predictions with HMMs

GenomeScan and HMMgene


Cédric Notredame (22/04/2015) Finding Genes In a Genome Cédric Notredame.

Documents

genome cdric notredame

gorf slide

gene slide

orfing slide

blast slide

way slide

operon slide

dna sequence slide