Folie 1

1
Description Recognition of Regulatory DNA
sequences
Rainer PudimatBioinformaticsFSU Jena
2
Titel dieser tollen Folie
Titel dieser tollen Folie
Titel dieser tollen Folie
Titel dieser tollen Folie
Titel dieser tollen Folie
Project Objectives
1

• Examining discriminating propertiesof regulatory
  DNA sequences

Architectural Proteins

• Applying alternative model approachesfor
  description and recognition

• Local Bayesian Networks as binding site models
• Global Decision Trees for explaining local
  matches

• Promoter Annotation Tool

Sp1
TATA box
introns exons
promoter
3
Titel dieser tollen Folie
Titel dieser tollen Folie
Titel dieser tollen Folie
Titel dieser tollen Folie
Project Objectives
The 2-Step Annotation Tool
2
Input Mammalian Upstream Sequence
E12
Sp1
MyoD
Score
MEF-2
TBP
SRF
Step 1 Scanning with Bayesian Networks
4
Titel dieser tollen Folie
Titel dieser tollen Folie
Titel dieser tollen Folie
Project Objectives
The 2-Step Annotation Tool
Bayesian Network Classifiers
3

• Directed Acyclic Graph (DAG)
• Nodes random variables
• Edges probabilistic dependencies
• Each node contains conditional probability
  distribution
• Network represents joint distribution of the
  variables

E12
A
D
B
C
P(a,b,c,d)
P(a,b,c,d) P(a)
P(a,b,c,d) P(a) P(ba)
P(a,b,c,d) P(a) P(ba) P(ca)
P(a,b,c,d) P(a) P(ba) P(ca) P(da,c)
Learning the Network
5
Titel dieser tollen Folie
Titel dieser tollen Folie
Project Objectives
The 2-Step Annotation Tool
Bayesian Network Classifiers
Common Scoring of Different Classifiers
4
E12
6
Titel dieser tollen Folie
Project Objectives
The 2-Step Annotation Tool
Bayesian Network Classifiers
Common Scoring of Different Classifiers
Application of Decision Trees
5
Should we play tennis ?
E12

• Each inner node corresponds to a test of some
  attribute
• Each outgoing edge to one possible value of
  that attribute
• Classification by moving down the tree
  according to the attribute values

Outlook
Sunny
Overcast
Rain
Y
Humidity
Wind
High
Normal
Strong
Weak
N
Y
N
Y
ltOutlSunny,TempHot,HumHigh,WindStronggt
Possible Questions for evaluating a putative
binding site
Composite Elements
Helical Parameters
Misc. Questions
pgtxpltx
gt 36lt 36
YESNO
MyoD present in neighbourhood ?
Helical Twist at -20 , 5 ?
Is in conservedregion ?
gt dlt d
YESNO
LiverMuscle
Distance ofnext MyoD ?
Periodic highbendable motifs ?
Tissue Inform.of Input ?

• "Worüber man nicht reden kann, darüber muss man
  schweigen."
• All tested knowledge has to be presentable in a
  numeric manner!

Ludwig Wittgenstein
7
Project Objectives
The 2-Step Annotation Tool
Bayesian Network Classifiers
Common Scoring of Different Classifiers
Application of Decision Trees
Issues on Adaquate Learning Samples
6

• To learn these global promoter characeristics
• Full annotated and experimental proved promoter
  sequences
• A number of negative samples
• A catalogue of valid questions
• Learning from False Positives
• could be useful to detect the "real" sufficient
  properties of regulatory sequences

E12
Building a Tree
Which test separatesbest TP from FP ?
Including the attributes of this test in the
question catalogues
Pleedings for a database for "proven false
positives"!
8
Project Objectives
The 2-Step Annotation Tool
Bayesian Network Classifiers
Common Scoring of Different Classifiers
Application of Decision Trees
Credits
7
E12

• Programming Data
• Java APIs jBNC, JavaBayes
• TRANSFAC
• TRRD

• Talk References
• Mitchel, Alberts, blablablabö
• ugwduhwhd
• euguw

• Cooperation
• Mitchel, Alberts, blablablabö
• ugwduhwhd
• euguw

• Contact
• rpudimat_at_informatik.uni-jena.de
• Tel,
• HP