View
215
Download
0
Embed Size (px)
Citation preview
Planning for Gene Regulatory Network Intervention
Daniel BryceArizona State University
Seungchan KimArizona State University &
Translational Genomics Research Institute
1/10/07 Bryce & Kim -- IJCAI-07 2
Prior Work Planning for Finding Pathways
S. Khan, K. Decker, W. Gillis, and C. Schmidt. “A multi-agent system-driven AI planning approach to biological pathway discovery.” In Proceedings of ICAPS’03, 2003.
Fifth International Planning Competition, 2006. Reasoning about change in cellular processes
N. Tran and C. Baral. “Issues in reasoning about interaction networks in cells: necessity of event ordering knowledge.” In Proceedings of AAAI’05, 2005.
Extracting and Expressing Transition Functions from Micro-array experiments, Markov chain analysis.
S. Kim, H. Li, E. Dougherty, N. Cao, Y. Chen, M. Bittner, and E. Suh. “Can Markov chain models mimic biological regulation?” Journal of Biological Systems, 10(4):337–357, 2002.I. Shmulevich, E. Dougherty, S. Kim, and W. Zhang.”Probabilistic boolean networks: a rule-based uncertainty model for gene regulatory networks.” Bioinformatics 18(2):261–274, 2002.
Non-AI work on planning interventions.A. Datta, A. Choudhary, M. Bittner, and E. Dougherty. “External control in Markovian genetic regulatory networks: the imperfect information case.” Bioinformatics, 20(6):924–930, 2004.
1/10/07 Bryce & Kim -- IJCAI-07 3
Gene Regulatory Networks (GRNs)
g1 g2
g3 g4
Gene Correlations
g1 g2 g3 g4
g1’ g2’ g3’ g4’
Dynamics Model
From: [Wuensche, PSB-98]
Cell Type(Phenotype, e.g., liver cell)
Tissue
Micro arrayData
Questions of interest: How does cancer occur? How can we prevent cancer? How do we kill specific cells? Can we control Differentiation?
e.g., Program stem cell to become Liver Cell
Can we change Phenotype? e.g., Revert liver cell to back to
stem cell, then differentiate to heart cell
1/10/07 Bryce & Kim -- IJCAI-07 4
Gene Regulatory Network Behavior
Steady States (normal)
Transient States(intermediate)
UndesirableState
Edge Thickness == Pr(s | s’)
CancerPhenotype
Extra cellular signals can effect the cell state transitions (e.g., Chemotherapy, Pharmaceuticals, and Stress)
Partial Observations of molecular components or physiology are available
1/10/07 Bryce & Kim -- IJCAI-07 5
GRN Intervention Planning
Datta et. al. Assumptions Synchronous Events Exact Representation Optimal Bounded Length
Plans Datta et. al. Approach
Enumerate Reachable Belief States
Dynamic Programming Our Approach
AI Planning AO* Search
Non-Intervention
Observation
Intervention
Observation
1/10/07 Bryce & Kim -- IJCAI-07 6
Evaluation
WNT5A GRN Highly active WNT5A
indicates proliferation of cancer
2 (non)interventions 2 variations: direct and
indirect control 2 observations 7 genes (binary valued)
Randomly Generated GRN 4 (non)interventions 2 observations 7 genes (binary valued)
Compare AI Planning with Datta et. al. Scaling horizon Sensitivity to
Reward Function Metric: Total Time
1/10/07 Bryce & Kim -- IJCAI-07 7
WNT5A GRN (from TGEN dataset)
Indirect Control Intevene Pirin gene Observe WNT5A gene
Direct Control Intervene WNT5A
gene Observe Pirin gene
Total Time
0
200
400
600
800
1000
1200
2 4 6 8 10 12
Horizon
Tim
e(s)
AO*
Datta
Total Time
0
1000
2000
2 4 6 8 10 12 14 16Horizon
Tim
e(s
)
AO*
Datta
1/10/07 Bryce & Kim -- IJCAI-07 8
Total Time
0
1000
2000
2 4 6 8 10 12
Horizon
Tim
e(s
)
-10-1110Datta
Random GRN (4 acts)
GoalReward(AO*)
Enumeration
AO* exploitsReward
Function forPruning
(ImprovedScalabilityIn Some Cases)
1/10/07 Bryce & Kim -- IJCAI-07 9
Assumptions Revisited
Finite Horizon Not all treatments require same length
Synchronous Change Actions overloaded to include GRN change
7 Genes and 1 intervention Within human comprehension
1/10/07 Bryce & Kim -- IJCAI-07 10
Indefinite or Finite Horizon?
Indefinite Horizon: If goal state is a steady state, then no need to plan more actions to meet a given horizon
1/10/07 Bryce & Kim -- IJCAI-07 11
Asynchronous Change
Decouple Intervention from Gene Regulatory Network Simulation Triggers (Tran and Baral, AAAI’05) Probabilistic Exogenous Events (Blythe,
UAI’94)
1/10/07 Bryce & Kim -- IJCAI-07 12
Larger GRNs
50-5000 genes More Interventions and Observations Representation:
ADD for transition relation blows up DBN is better, but exact inference can be costly Extensions of Thrun’s MC-POMDP’s, sample based
representation, is in the right direction
Search Heuristics: McLUG: Planning Graphs with Probabilistic Actions
1/10/07 Bryce & Kim -- IJCAI-07 13
Conclusion
Off-the-shelf AI planning improves upon state of the art in Intervention Problems
Future Research Needed: Scaling
Indefinite Horizon Extra Actions and Observations Sample-based Representation Search Heuristics
Modeling Asynchronous Probabilistic Change
Plan Explanation
1/10/07 Bryce & Kim -- IJCAI-07 15
Empirical Comparison
Total Time and Expanded Nodes
Better in all Cases
With no heuristicsSearch performance
Correlates with RewardFunction
Datta Enumeration
AO*
1/10/07 Bryce & Kim -- IJCAI-07 20
Computational Biology
Bioinformatics Knowledge Discovery & Data-mining Manage and Analyze Biological Data
Systems Biology Simulation Model Dynamic Systems
1/10/07 Bryce & Kim -- IJCAI-07 21
Representing State Distributions
2.
2.
35.
25.
21
2
_
1
21
_2
_
1
_
gg
gg
gg
gg
Explicit Vectorg1
g2 g2
.2 .25 .35
Algebraic Decision Diagram
1/10/07 Bryce & Kim -- IJCAI-07 22
Representing State Distributions
2.
2.
35.
25.
21
2
_
1
21
_2
_
1
_
gg
gg
gg
gg
Explicit Vectorg1
g2
.2 .25 .35
Algebraic Decision Diagram
1/10/07 Bryce & Kim -- IJCAI-07 23
Representing Probabilistic Actions
0001
08.2.0
3.06.1.
05.3.2.
21
2
_
1
21
_2
_
1
_
2'1'2'_
1'2'1
_'2
_'1'
_
gg
gg
gg
gg
gggggggg g1
g’1 g’1
g’2 g’2 g’2 g’2
g2 g2 g2 g2
g’2 g’2 g’2 g’2
0 .1 .2 .3 .5 .6 .8 1
Explicit Transition Matrix Algebraic Decision Diagram
1/10/07 Bryce & Kim -- IJCAI-07 24
Representing Probabilistic Actions
g1
g’1 g’1
g’2 g’2 g’2 g’2
g2 g2 g2 g2
g’2 g’2 g’2
0 .1 .2 .3 .5 .6 .8 1
0001
08.2.0
3.06.1.
05.3.2.
21
2
_
1
21
_2
_
1
_
2'1'2'_
1'2'1
_'2
_'1'
_
gg
gg
gg
gg
gggggggg
Explicit Transition Matrix Algebraic Decision Diagram
1/10/07 Bryce & Kim -- IJCAI-07 25
Modeling Network Dynamics
?g
?g1 ?g2 ?g3 ?g4 0 1
(- (influence1 ?g1 ?g2 ?g) (noise))
(- (influence2 ?g3 ?g4 ?g) (noise))
(noise) (noise)
?g1 ?g2 ?g(not (up-regulated ?g1)) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g zz)(not (up-regulated ?g1)) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g zo) (up-regulated ?g1) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g oz) (up-regulated ?g1) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g oo)
?g1 ?g2 ?g(not (up-regulated ?g1)) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g zz)(not (up-regulated ?g1)) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g zo) (up-regulated ?g1) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g oz) (up-regulated ?g1) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g oo)
(predicts ?g1 ?g2 ?g) (predicts ?g3 ?g4 ?g)
1/10/07 Bryce & Kim -- IJCAI-07 26
Network Dynamics Encoding <dynamics>(forall (?g ?g1 ?g2 ?g3 ?g4 - gene) ;;constraint for grounding that binds only those genes ?g1 - ?g4 that ;;predict ?g. External control actions add predicates to the ;;antecedent below so that ?g does not bind to controlled genes. (when (and (predicts1 ?g1 ?g2 ?g) (predicts2 ?g3 ?g4 ?g)) (probabilistic (- (influence1 ?g1 ?g2 ?g) (noise)) ;;predictor 1 probability (and (when (or ;;conditions to set ?g up (and (not (up-regulated ?g1)) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g zz)) (and (not (up-regulated ?g1)) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g zo)) (and (up-regulated ?g1) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g oz)) (and (up-regulated ?g1) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g oo)) ) (up-regulated ?g)) ;;set ?g up (when (or ;;conditions to set ?g down (and (not (up-regulated ?g1)) (not (up-regulated ?g2)) (not (pred-fn ?g1 ?g2 ?g zz))) (and (not (up-regulated ?g1)) (up-regulated ?g2) (not (pred-fn ?g1 ?g2 ?g zo))) (and (up-regulated ?g1) (not (up-regulated ?g2)) (not (pred-fn ?g1 ?g2 ?g oz))) (and (up-regulated ?g1) (up-regulated ?g2) (not (pred-fn ?g1 ?g2 ?g oo))) ) (not (up-regulated ?g))) ;;set ?g down ) (- (influence2 ?g3 ?g4 ?g) (noise)) ;;predictor 2 probability (and [...]) ;;predictor 2, similar to predictor 1 (noise) (up-regulated ?g) ;;noise to set ?g up (noise) (not (up-regulated ?g)) ;;noise to set ?g down ) ))
conditions to up-regulate
with predictor1
Conditions to down-regulate
up-regulatewith predictor1
down-regulatewith predictor1
probabilityOf using
predictor1
Binding constraints
Rules for predictor2
and noise
Bind all genes to variables
1/10/07 Bryce & Kim -- IJCAI-07 27
Network Parameters
s100p
wnt5a ret2 stc2 ret2 0 1
.68 .30 .01 .01
?g1 ?g2 ?g(not (up-regulated ?g1)) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g zz)(not (up-regulated ?g1)) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g zo) (up-regulated ?g1) (not (up-regulated ?g2)) (pred-fn ?g1 ?g2 ?g oz) (up-regulated ?g1) (up-regulated ?g2) (pred-fn ?g1 ?g2 ?g oo)
(predicts wnt5a ret2 s100p) (predicts stc2 ret2 s100p)
wnt5a ret2 s100p(not (up-regulated wnt5a)) (not (up-regulated ret2)) (not (up-regulated s100p))(not (up-regulated wnt5a)) (up-regulated ret2) (up-regulated s100p) (up-regulated wnt5a) (not (up-regulated ret2)) (up-regulated s100p) (up-regulated wnt5a) (up-regulated ret2) (up-regulated s100p)
1/10/07 Bryce & Kim -- IJCAI-07 28
Predictor Encoding
(= (noise) .01)
;s100p predictor1(predicts1 wnt5a ret2 s100p)
(pred-fn wnt5a ret2 s100p zo) ;1 (pred-fn wnt5a ret2 s100p oz) ;1 (pred-fn wnt5a ret2 s100p oo) ;1
(= (influence1 wnt5a ret2 s100p) .69)
;s100p predictor2(predicts2 stc2 ret2 s100p)
(pred-fn stc2 ret2 s100p oz) ;1 (pred-fn stc2 ret2 s100p oo) ;1
(= (influence2 stc2 ret2 s100p) .31)
stc2 ret2 s100p(not (up-regulated stc2)) (not (up-regulated ret2)) (not (up-regulated s100p))(not (up-regulated stc2)) (up-regulated ret2) (not (up-regulated s100p)) (up-regulated stc2) (not (up-regulated ret2)) (up-regulated s100p) (up-regulated stc2) (up-regulated ret2) (up-regulated s100p)
wnt5a ret2 s100p(not (up-regulated wnt5a)) (not (up-regulated ret2)) (not (up-regulated s100p))(not (up-regulated wnt5a)) (up-regulated ret2) (up-regulated s100p) (up-regulated wnt5a) (not (up-regulated ret2)) (up-regulated s100p) (up-regulated wnt5a) (up-regulated ret2) (up-regulated s100p)
1/10/07 Bryce & Kim -- IJCAI-07 29
Control Encoding <control>, perfect/partial obeservation
(:action down-regulate :parameters (?gr ?go - gene) :precondition (and (observed ?go)
(controlled ?gr) (started)) :effect (and (decrease (reward) 1)
(when (up-regulated ?gr) (not (up-regulated ?gr))) <dynamics + (not (= ?g ?gr))> ) :observation (
((up-regulated ?go) (up-regulated ?go) 1) ((not (up-regulated ?go)) (not (up-regulated ?go)) 1)
))
(when (up-regulated ?gr) (probabilistic .75 (not (up-regulated ?gr))))
Could be better model!?