prism - win.tue.nlevink/education/2if35/pdf/2if35-05a.pdf12 / department of mathematics and computer...

45
12 / department of mathematics and computer science 1/48 PRISM 2IF35 – week 5

Upload: vuxuyen

Post on 23-Aug-2019

213 views

Category:

Documents


0 download

TRANSCRIPT

12

/ department of mathematics and computer science 1/48

PRISM

2IF35 – week 5

12

/ department of mathematics and computer science 3/48

properties of discrete-time Markov chains

I until:

P ≥ 0.92(

¬illegal U goal

)I bounded until:

P≥ 0.92(

¬illegal U≤137 goal

)

I eventually:

P≥ 0.92(

true U

=137

goal

)

P≥ 0.92(

=137

goal

)

I always:

P≥ 0.92(

� goal

)

P≤ 0.08(

¬♦¬goal

)

I nested:

P≥ 0.92(

¬illegal U

P≥ 0.9999

(�

=13

goal

)

)

12

/ department of mathematics and computer science 3/48

properties of discrete-time Markov chains

I until:

P≥ 0.92(

¬illegal U goal

)

I bounded until:

P≥ 0.92(

¬illegal U≤137 goal

)

I eventually:

P≥ 0.92(

true U=137 goal

)

P≥ 0.92(

♦=137 goal

)

I always:

P≥ 0.92(

� goal

)

P≤ 0.08(

¬♦¬goal

)

I nested:

P≥ 0.92(

¬illegal U

P≥ 0.9999

(�=13goal

)

)

12

/ department of mathematics and computer science 3/48

properties of discrete-time Markov chains

I until: P≥ 0.92(¬illegal U goal

)I bounded until: P≥ 0.92

(¬illegal U≤137 goal

)I eventually: P≥ 0.92

(true U=137 goal

)≡ P≥ 0.92

(♦=137 goal

)I always: P≥ 0.92

(� goal

)≡ P≤ 0.08

(¬♦¬goal

)I nested: P≥ 0.92

(¬illegal U P≥ 0.9999

(�=13goal

) )

12

/ department of mathematics and computer science 5/48

properties of continuous-time Markov chains

I P≥ 0.92(¬illegal U≤133.4 goal

)I P≥ 0.92

(¬illegal U[61.1,133.4] goal

)

12

/ department of mathematics and computer science 6/48

Prism: a probabilistic modelchecker

I discrete-time and continuous-time Markov chainsI reactive module-type of modelling languageI specifications in PCTL and CSLI see http://www.prismmodelchecker.org

Kwiatkowska et al., Proc. QEST (2004)

12

/ department of mathematics and computer science 7/48

a kitchen-salt example

Na++ Cl−

K fKb

NaCl

12

/ department of mathematics and computer science 8/48

a Prism model

module Solution

Na=0, Cl=0, NaCl=50;

[ ] Na > 0 & Cl > 0 -> Kf∗Na∗Cl :

Na’=Na-1 & Cl’=Cl-1 & NaCl’=NaCl+1;

[ ] NaCl > 0 -> Kb∗NaCl :

Na’=Na+1 & Cl’=Cl+1 & NaCl’=NaCl-1;

endmodule

12

/ department of mathematics and computer science 9/48

experiments

percentage NaCl

12

/ department of mathematics and computer science 10/48

CSL properties

I ’init’ ⇒ �0.98 (Na = Cl ) (valid for Kf=0.01,Kb=1)

levels of Na and Cl are always the same

I ’init’ ⇒ ♦ (Na > NaCl ) (valid for Kf=0.01,Kb=1)

at some point in time the Na level will exceed that of NaCl

I ’init’ ⇒ ♦≤10 (Na > NaCl ) (not valid for Kf=0.01,Kb=1)

the same but within 10 time units

I ’init’ ⇒ ♦≤20 (Na > NaCl ) (valid for Kf=0.01,Kb=1)

again the same but now within 20 time units

12

/ department of mathematics and computer science 12/48

Example: two process mutual exclusion

nondeterministic

module M1x : [0..2] init 0;

[] x=0 -> 0.8:(x’=0) + 0.2:(x’=1);[] x=1 & y!=2 -> (x’=2);[] x=2 -> 0.5:(x’=2) + 0.5:(x’=0);

endmodule

module M2y : [0..2] init 0;

[] y=0 -> 0.8:(y’=0) + 0.2:(y’=1);[] y=1 & x!=2 -> (y’=2);[] y=2 -> 0.5:(y’=2) + 0.5:(y’=0);

endmodule

12

/ department of mathematics and computer science 14/48

Example: N-place queue and server

ctmc

const int N = 10;const double mu = 1/10, lambda = 1/2, gamma = 1/3;

module queueq : [0..N];

[] q<N -> mu:(q’=q+1);[] q=N -> mu:(q’=q);[serve] q>0 -> lambda:(q’=q-1);

endmodule

module servers : [0..1];

[serve] s=0 -> 1:(s’=1);[] s=1 -> gamma:(s’=0);

endmodule

12

/ department of mathematics and computer science 15/48

circadian clock: activator and repressor

12

/ department of mathematics and computer science 17/48

circadian clock

ctmcconst int N=1000000;const double k;

module activator_gene

da : [0..1] init 1;da_a : [0..1];

[transc_da] da=1 -> da : true;[transc_da_a] da_a=1 -> da_a : true;[bind_a] da=1 & da_a=0 -> da : (da’=0) & (da_a’=1);[rel_a] da=0 & da_a=1 -> da_a : (da’=1) & (da_a’=0);

endmodule

12

/ department of mathematics and computer science 19/48

activator modules

module activator_mRNAma : [0..N]; // amount of activator mRNA

[transc_da] ma<N -> ma’=ma+1;[transc_da_a] ma<N -> ma’=ma+1;[transl_a] ma>0 -> ma : true;[deg_ma] ma>0 -> ma : ma’=ma-1;

endmodule

module activator_proteina : [0..N]; // amount of activator protein

[transl_a] a<N -> a’=a+1;[bind_a] a>0 -> a : a’=a-1;[bind_r] a>0 -> a : a’=a-1;[rel_a] a<N -> a’=a+1;[rel_r] a<N -> a’=a+1;[deg_a] a>0 -> a : a’=a-1;[deactive] a>0 -> a : a’=a-1;

endmodule

12

/ department of mathematics and computer science 21/48

repressor modules// rename activator gene modulemodule repressor_gene =

activator_gene[da=dr, da_a=dr_a, bind_a=bind_r, rel_a=rel_r,transc_da=transc_dr, transc_da_a=transc_dr_a]

endmodule

// rename activator mRNA modulemodule repressor_mRNA =

activator_mRNA[ma=mr, transl_a=transl_r, transc_da=transc_dr,transc_da_a=transc_dr_a, deg_ma=deg_mr]

endmodule

module repressor_proteinr : [0..N];

[transl_r] r<N -> (r’=r+1);[deg_r] r>0 -> r : (r’=r-1);[deactive] r>0 -> r : (r’=r-1);[deg_c] r<N -> (r’=r+1);

endmodule

12

/ department of mathematics and computer science 23/48

complexes and rates

module inactive_proteinc : [0..N]; // amount of inactivated protein

[deactive] c<N -> (c’=c+1);[deg_c] c>0 -> c : (c’=c-1);

endmodule

module rates[transc_da] true -> k*50 : true;[transc_da_a] true -> k*500 : true;[transc_dr] true -> k*0.01 : true;[transc_dr_a] true -> k*50 : true;[transl_a] true -> 50 : true;[transl_r] true -> 5 : true;[bind_a] true -> 1 : true;[bind_r] true -> 1 : true;[deactive] true -> 2 : true;...

endmodule

12

/ department of mathematics and computer science 25/48

reward structure

const double T;

rewards "activated_protein"true : a;endrewards

12

/ department of mathematics and computer science 26/48

simulation

single run

average over 20 runs

number of activator proteins over time

12

/ department of mathematics and computer science 26/48

simulation

single run average over 20 runs

number of activator proteins over time

12

/ department of mathematics and computer science 27/48

transcription factor

activator level

12

/ department of mathematics and computer science 28/48

aa-tRNA competition

I Fluitt, Pienaar and ViljoenComputational Biology and Chemistry 31 (2007) 335–346

I building on work of Rodnina c.s.codon CUC with Leu cognate and Phe near-cognate

I simplifications:• constant tRNA pool• no ribosome-ribosome interaction• constants not measured are not rate limiting

I simulation effort: correlation with competition

12

/ department of mathematics and computer science 29/48

transcription & translation

protein

mRNA

DNA

mRNA: string of codons protein: string of amino acids

12

/ department of mathematics and computer science 30/48

the genetic code

12

/ department of mathematics and computer science 31/48

codon & anticodon

anticodon

aminoacid

tRNA

codonmRNA

E. coli: 64 codons, 48 aa-tRNAs/anticodons, 20 amino acids

12

/ department of mathematics and computer science 32/48

aa-tRNA

specifity aa-tRNA by anticodon

12

/ department of mathematics and computer science 33/48

ribosome translating mRNA into protein

12

/ department of mathematics and computer science 34/48

aa-tRNA competition

I aa-tRNAs compete for matching with codon

I aa-tRNAs classified ascognate, near-cognate, or non-cognate

I fidilty of translation

• insertion errorwrong amino acid added to nascent chain

• insertion timetime for successful elongation, amount of delay

12

/ department of mathematics and computer science 35/48

peptidyl elongation I

12

/ department of mathematics and computer science 36/48

peptidyl elongation II

12

/ department of mathematics and computer science 37/48

probabilistic automaton

0.23 / 80167 / 46

60 / FAST

260 / 0.40190

85

1

0

2 3

5

4 6 8arrival

FAST

FAST

cognate and near-cognate rates differ

12

/ department of mathematics and computer science 37/48

probabilistic automaton

0.23 / 80167 / 46

60 / FAST

260 / 0.40190

85

1

0

2 3

5

4 6 8arrival

FAST

FAST

cognate and near-cognate rates differ

arrival rate depends on concentration

12

/ department of mathematics and computer science 39/48

Prism representation

// initial binding[ ] (s=1) -> k1f * c_cogn : (s’=2) & (cogn’=true) ;[ ] (s=1) -> k1f * c_near : (s’=2) & (near’=true) ;[ ] (s=1) -> k1f * c_nonc : (s’=2) & (nonc’=true) ;

// dissociation[ ] (s=2) & ( cogn | near ) -> k2b :

(s’=0) & (cogn’=false) & (near’=false) ;[ ] (s=2) & nonc -> k2bx : (s’=1) & (nonc’=false) ;

// codon recognition[ ] (s=2) & ( cogn | near ) -> k2f : (s’=3) ;[ ] (s=3) & cogn -> k3bc : (s’=2) ;[ ] (s=3) & near -> k3bn : (s’=2) ;

12

/ department of mathematics and computer science 41/48

Prism representation (cont.)

// rejection[ ] (s=4) & cogn -> k4rc : (s’=5) & (cogn’=false) ;[ ] (s=4) & near -> k4rn : (s’=5) & (near’=false) ;

// accommodation, peptidyl transfer[ ] (s=4) & cogn -> k4fc : (s’=6) ;[ ] (s=4) & near -> k4fn : (s’=6) ;

// no entrance, re-entrance at state 1[ ] (s=0) -> FAST : (s’=1) ;// rejection, re-entrance at state 1[ ] (s=5) -> FAST : (s’=1) ;// elongation[ ] (s=8) -> FAST : (s’=8) ;

12

/ department of mathematics and computer science 42/48

matching for E. coliEscherichia coli has for

I codon AAU coding amino acid Asparagine has

• 1 cognate anticodons GUU• 1 near-cognate anticodon UUU• 44 non-cognate anticodons

I codon CUC coding amino acid Leucine has

• 1 cognate anticodon GAG• 4 near-cognate anticodons GUG, CAG, UAG, GGG• 41 non-cognate anticodons

I codon AUC coding amino acid Isoleucine has

• 1 cognate anticodon GAG• 8 near-cognate anticodons GUU, CAU, . . .• 37 non-cognate anticodons

12

/ department of mathematics and computer science 43/48

insertion error

conditional cognate probability for success

P=?[ (s!=0 & s!=5)U(s=8) {(s=2)&cogn} ]

conditional near-cognate probability for success

P=?[ (s!=0 & s!=5)U(s=8) {(s=2)&near} ]

0.23 / 80167 / 46

60 / FAST

260 / 0.40190

85

1

0

2 3

5

4 6 8arrival

FAST

FAST

success for a cognate pcs = 0.508

success for a near-cognate pns = 0.484 · 10−4

12

/ department of mathematics and computer science 44/48

insertion error (cont.)

cognate probability for success: proper elongation

P=?[ true U((s=8)&cogn) ]

near-cognate probability for success: insertion error

P=?[ true U((s=8)&near) ]

0.23 / 80167 / 46

60 / FAST

260 / 0.40190

85

1

0

2 3

5

4 6 8arrival

FAST

FAST

error probability depends on concentrations

12

/ department of mathematics and computer science 45/48

insertion error (cont.)

strong correlation insertion errors and ratio near-cognates/cognates

12

/ department of mathematics and computer science 45/48

insertion error (cont.)

strong correlation insertion errors and ratio near-cognates/cognates

probability for erroneous insertion

ratio n

ear−

cognate

/ c

ognate

12

/ department of mathematics and computer science 45/48

insertion error (cont.)

strong correlation insertion errors and ratio near-cognates/cognates

P(error) =pn

s · (fn/tot)

pcs · (fc/tot) + pn

s · (fn/tot)

≈pn

s · fn

pcs · fc

∼fn

fc

12

/ department of mathematics and computer science 46/48

insertion times

average exit times

e.g., dissociation or rejection of near-cognate

R=?[F((s=0)|(s=5)){(s=2)&near}]

0.23 / 80260 / 0.40

1901

0

2 3

5

4 6 8arrival

FAST

FAST

60+67 FAST65 /

success/failure for cognate T cs = 0.03182, T c

f = 9.342 · 10−3

success/failure for near-cognate T ns = 3.251, T n

f = 0.3914

12

/ department of mathematics and computer science 47/48

insertion times (cont.)

strong correlation insertion times and ratio near-cognates/cognates

12

/ department of mathematics and computer science 47/48

insertion times (cont.)

strong correlation insertion times and ratio near-cognates/cognates

average insertion time

ratio (

pseudo−

cognate

+ n

ear−

cognate

) / cognate

12

/ department of mathematics and computer science 47/48

insertion times (cont.)

strong correlation insertion times and ratio near-cognates/cognates

Tins =∑

i=0 (pcf )

ipcs · ((delay for i + 1 cognates) + T c

s )

=∑

i=0 (pcf )

ipcs ·

(i · (T c

f +fn

fcT n

f +fx

fcT x

f ) + T cs

)≈

fn

fcpc

s Tnf∑

i=0 i (pcf )

i

∼fn

fc

12

/ department of mathematics and computer science 48/48

Conclusion

I Stochastic automata fit nicely• Prism and CSL adequate

I Model of translation confirms experimentsI Piecewise analysis to support explanation

• Better insight compared to simulation