Transcript ppt - Sol Genomics Network

A Bioinformatic Framework to
Unravel the Secrets of the Tomato
Genome
15 January 2006, PAG XIV SanDiego
Rémy Bruggmann, MIPS/IBI, GSF
Outline
 Introduction
 Data management
 Annotation
 Training/Test gene set
 Summary
MIPS´ look at the Green Side of Life
– genome projects and database activities –
Arabidopsis thaliana
Arabidopsis lyrata *
Capsella rubella *
Maize
Rice
Medicago
Lotus
Solanum lycopersicum
MIPS´ look at the Green Side of Life
– genome projects and database activities –
 Need to streamline and unify databases as well as
analytical schemas and operation routines
 Strong synergism and very robust
 Risk to loose flexibility and „custom tailor“ attractiveness
 Awareness that not every genome and every community
„is just the same“
From Center Centric Strategies to
distributed Approaches
Typically, genome projects undergo particular
phases:
 Sequenced BACs are annotated
 Gene models are published to the community
 Potentially generates competition rather than
collaboration among groups
From Center Centric Strategies to
distributed Approaches
Consequences can be:
 underlying analytical procedures are not always tested,
trained and evaluated
 Between groups more or less pronounced differences exist
--> differing, contradicting and confliciting data
Aim of all groups:
„information enriched high quality genome
backbone to address genome scale biological
questions“
From Center Centric Strategies to
distributed Approaches
An example ...
 International Medicago Genome Annotation Group
 Consists of groups participating either in the International or
the European Medicago Genome Initiative annotation/
bioinformatics programs
 Agreement on common annotation standards, data
exchange formats and naming conventions
 Aims to produce and provide unified high-quality Medicago
data set
From Center Centric Strategies to
distributed Approaches
Advantages of sharing efforts in genome
annotation within a common annotation pipeline
From Center Centric Strategies to
distributed Approaches
 prevents from:
(i) duplicating efforts
(ii) conflicts resulted from different
annotation “standards”
 ensures high-quality annotation standards
 ensures common (gene) naming  common dataset
 Integrates and profits from knowledge and expertise
of the individual groups
Data management
All data should be organized in a
genome database
Wishlist for a modern genome db









Complete
Comprehensive
Up-to-date
Integrated
User interface
Application interface
State-of-the-art automatic analysis
Adaptable
Cross-genome comparison
…low cost, low manpower...
PlantsDB Philosophy
 Plants Genome Resource: provides and integrates
sequence data from European plant sequencing
consortia along with publically available data from the
international initiative
 Plants DB communicates bioinformatic analysis data
(visualization, genetic elements, structural data,
ontologies, domains...; BLAST, browse and
search,…comparative analysis)
 Integration: provides a distributed network to integrate
and retrieve data from heterogenous resources using
BioMOBY (connection to other plant DBs, PlaNET)
Preliminary Annotation Pipeline
Towards a preliminary annotation
Repeat Detection
Repeat Ontology
Repeat
Database
RepeatMasker
Masked sequences
Gene prediction
Repeat annotation
GAME
XML
Gene Prediction
External Databases
Gene prediction programs
EST DB
EST
Assemblies
► GenomeThreader
► FGenesH++/ProtMap
Document of
computational
results
► GeneMarkHMM
GAME
XML
Protein DB
e.g.
SwissProt
Manual annotation in
Apollo Genome Viewer
Web Access
Gbrowse
PlantsDB
First Results
Repeat Masker
[%]
 5.8 MB analysed (48 BACs)
25
 ~ 6.7 % repetitive elements
(<0.2% - 23% per bac)
20
 ~ 1 min/100 kb
15
Repeat content
10
whole genome (euchromatic part):
5
~ 2 days
0
BACs
State: December 2005
Preliminary Results
Comparison of different gene finders
ab initio predictions
EST/TC
GeneMark
FGeneSH
EST/TC
ab initio predictions
ab initio predictions
 FGeneSH++ and GeneMarkHMM often generate
incomplete or wrong gene models at the moment
 There are no matrices available that are trained for tomato

Tomato matrices will increase prediction quality
dramatically

Collection of annotated high quality genes for a
training/test set for EuGene, FGeneSH,
GeneMarkHMM, ...
Training/Test Gene Set
How can we get a training/test set?

Map available tomato cDNA/ESTs to the BACs
(use only high confident matches)

Link experimental data to the genemodels

Use this gene set for ab initio gene finder training
GenomeThreader
GenomeThreader used for EST/cDNA-Mapping:
 similarity-based approach:
EST/Proteins used to predict gene structure via optimal
spliced alignments
 Offers many options (full user control)
 incremental updates (avoids a lot of duplicated
computations)
 Improved GeneSeqer
GenomeThreader - calculations
DB
Entries Size [MB]
Calc time/100kb [s]
Whole Genome
Tomato
32401
27
27 s
MicroTom
26363
21
22 s
Potato
38239
34
23 s
Tobacco
28661
20
39 s
Arabidopsis cDNAs
31939
45
10 s
0.3 days
404822
311
170 s
4.3 days
15639
21
8s
0.2 days
Uni_trembl Plants
185564
74
38 s
1.0 day
Uniprot_swissprot
181571
82
8s
0.2 days
Nonred
1675230
662
437 s
11.1 days
Total
2834224
1433
14 min
22 days
Dicots
rice cds
~ 2.8 days
(single CPU, euchromatic part)
Example
Tobacco
Potato
Microtom
Tomato
Examples - UK
Example
Number of high quality genes
# genes
 Number of genes: 164
(covered completely by
cDNA/ESTs)
10
8
6
 ~3.4 genes/BAC
(range: 0 - 9 genes/BAC)
4
2
 These genes can be used
to train gene finders
(Only very good alignments considered)
0
BAC
Gene Finder
Which program can be trained for tomato?
One possibility is EuGene (VIB Gent)
- performed well e.g. for Arabidopsis and Medicago
- available as soon as test/training gene set is large
enough
EuGene - overview
Plugins
Statistical contents
DNA Markov
AA Markov
Splice sites
NetGene2
GeneSplicer
SpliceMachine
SplicePredictor
Start sites
SpliceMachine
NetStart
ATRPred
Similarities
EST similarities
Protein similarities
FL cDNA
Repeats
Exon conservation
Plugin
training
Optimize
plugin
Test
combination
Needs
Needs
Needs
one
one
one
dataset
dataset
new
dataset
TRAINING
OPTIM
TEST
EuGene
 First round training:
- 500 high quality tomato genes
- statistical models on codon usage and splice sites of
Arabidopsis will be used
 Second round training:
- 2000 high quality tomato genes
- Build a tomato-only version of EuGene
Approx. 150 BACs needed for first round training
Current state of sequenced BACs
Total number of BACs:
- unfinished:
71
- finished:
87
- available:
52
Summary
 ab initio gene finders are not yet calibrated to tomato
 Need of a test/training gene set to calibrate the gene finders
 We need another 100 BACs to get enough genes for a first
round training of EuGene
 GenomeThreader produces only good alignments with ESTs
from SOL-species (Tomato, Potato, Tobacco)
 More repeats will be detected (will be included in
RepeatMasker Library)
Acknowledgments
Automated annotation
MIPS
Heidrun Gundlach
Georg Haberer
Manuel Spannagl
Klaus F.X. Mayer
Manual Annotation/Curation/Web-site
(Chromosome 4)
Imperial College
Daniel Buchan
James Abbot
Sarah Butcher
Gerard Bishop
Sequencing & Assembly
(Chromosome 4)
Sanger Institute
Christine Nicholson
Sean Humphray
MPIZ Köln
Heiko Schoof
EuGene
VIB Gent
Stephane Rombauts
GenomeThreader
University of Hamburg
Gordon Gremme
Stefan Kurtz
Volker Brendel
A Bioinformatic Framework to
Unravel the Secrets of the Tomato
Genome
15 January 2006, PAG XIV SanDiego
Rémy Bruggmann, MIPS/IBI, GSF