uncovering signaling transduction networks from ppi network by inductive logic programming woo-hyuk...
TRANSCRIPT
Uncovering Signaling Transduction Networks from PPI network by Inductive Logic
Programming
Woo-Hyuk Jang2009. 3. 20
Contents
• Introduction
• Method– ILP system (ALEPH)– ILP modeling example (Marriage Case)
• ILP modeling of STP
• Challenges and Future Work
Introduction
• Most of Signaling Transduction Network (STN) prediction methods follow the sequences, 1) making integrative PPIs, 2) Finding rules from STN, 3) discovering STN components from PPIs.
• In addition, these methods generally adopt probabilistic model in each sequence.
• However,– Accumulation of even small noise may lead to big prediction inac
curacy.– Probabilistic model cannot provide biological explanation of the re
sults.
Related Work
• Steffen, et. al. (2002)– Integrating PPI and microarray data– Netsearch algorithm
• Yin Liu and Hongyu Zhao (2004)– Ordering proteins when all components of STN are already known.
• Jacob Scott, et. al. (2006)– A variant of the color coding algorithm – Yeast PPI
• Gurkan Bebek and Jiong Yang (2007)– Extract functional patterns from STP– PathFinder
• Xing-Ming Zhao, et. al. (2008)– PPI + gene expression profile– Integer linear programming
New Approach
ILP STPPPINetwork
PreSPI
Functional Patterns
Corrects True Negative, False Positive path
Features
Reference
Induced Rules
Method
• Inductive Logic Programming (ILP)– Programs that “generalize”
– Programs that follow the Specific General idea
Molecular structure of toxic and non-toxic chemicals, other props …
Chemical is toxic if it has a ring connected to… and a C atom in… …
Method
• Inductive Logic Programming (ILP)– A powerful representation language
• Express complex relationships easily
– Easy to provide background information• Including other analysis methods like regression etc
– We can easily integrate diverse features and their relations that may affect to PPI in STP
A Learning Engine for Proposing Hypotheses(ALEPH)
• ILP system that follows a very simple procedure that can be described in 4 steps:– 1. Select example– 2. Build most-specific-clause– 3. Search– 4. Remove redundant
• Background knowledge (*.b), Positive example (*.f), Negative example (*.n)
ILP modeling example Case 1 (Marriage)
• There are some features that have driven this couple to fall
in love
Propertyoccupation
Pos. in brothers personality
. . .
Case 1 (Marriage)
• Mode DeclarationsMale1
Property 10억
Occupation 의사
Pos. in. brothers
Youngest
Personality Very good
Male2
Property 100억
Occupation 사업가
Pos. in. brothers
Eldest
Personality good
Male3
Property 0
Occupation 의사
Pos. in. brothers
Youngest
Personality Very bad
Male4
Property 0
Occupation 없음
Pos. in. brothers
Eldest
Personality Very good
Female1
Property 5억
Occupation 의사
Pos. in. brothers
Youngest
Personality Very good
Female2
Property 1억
Occupation 학생
Pos. in. brothers
middle
Personality Very good
Female3
Property 10억
Occupation 의사
Pos. in. brothers
Youngest
Personality Very bad
Female4
Property 0
Occupation 없음
Pos. in. brothers
Middle
Personality Very good
Background Knowledge
Positive & Negative Example
• Positive Example
Person Person
Male1 Female1
Male2 Female2
Male3 Female3
Male4 Female4
• Negative Example
Person Person
Male1 Female4
Male2 Female3
Male3 Female2
Male4 Female1
Case 1 (Marriage)
• Mode DeclarationsMale1
Property 10억
Occupation 의사
Pos. in. brothers
Youngest
Personality Very good
Male2
Property 100억
Occupation 사업가
Pos. in. brothers
Eldest
Personality good
Male3
Property 0
Occupation 의사
Pos. in. brothers
Youngest
Personality Very bad
Male4
Property 0
Occupation 없음
Pos. in. brothers
Eldest
Personality Very good
Female1
Property 5억
Occupation 의사
Pos. in. brothers
Youngest
Personality Very good
Female2
Property 1억
Occupation 학생
Pos. in. brothers
middle
Personality Very good
Female3
Property 10억
Occupation 의사
Pos. in. brothers
Youngest
Personality Very bad
Female4
Property 0
Occupation 없음
Pos. in. brothers
Middle
Personality Very good
Approach Again
ILP STPPPINetwork
PreSPI
Functional Patterns
Corrects True Negative, False Positive path
Features
Reference
Induced Rules
segmentsevaluation
ILP modeling on STP
• We can rewrite STP as a sequence of protein pair
STE2/3 Gpa1 Ste4/18 Cdc42 Ste20
Pheromone response in MAPK STP
Interaction(STE2/3, GPA1).
Interaction(GPA1, STE4/18).
Interaction(STE4/18, CDC42).
Couple(person, person)
W_property(male1,10)Go_of(STE2/3, GOXXXXX).
GO_of(GPA1, GOXXXXX).
GO_of(STE4/18, GOXXXXX).
Feature Selection
• T. P. Nguyen and T. B. Ho, “Discovering Signal Transduction Networks Using Signaling Domain-Domain Interaction”, 2006, Genome Informatics.
Challenges and Future Work
• Refined feature selection
• Build parser for each biological DB
• Mode declaration– Build determination predicates
• Evaluation problem– Induced rule from MAPK extracting segments from PPI
compared to MAPK ???