Download presentation
Presentation is loading. Please wait.
Published byDerrick Cross Modified over 9 years ago
1
Stochastic modeling of molecular reaction networks Daniel Forger University of Michigan
2
Let’s begin with a simple genetic network
3
We can list the basic reaction rates and stochiometry numsites = total # of sites on a gene, G = # sites bound M = mRNA, Po = unmodified protein, Pt = modified protein Transcription trans or 0+M Translation tl*M+Po Protein Modification conv*Po-Po, +Pt M degradation degM*M-M Po degradationdegPo*Po-Po Pt degradationdegPt*Pt-Pt Binding to DNAbin(numsites - G)*Pt -Pt, +G Unbinding to DNAunbin*G-G
4
We normally track concentration Let’s track # molecules instead Let M, Po, Pt be # molecules First order rate constants (tl, unbin, conv, degM, degPo and degPt) have units 1/time and stay constant Zero order rate constant (trans) has units conc/time, so multiply it by volume 2nd order rate constant (bin) has units 1/(conc*time), so divide it by volume
5
numsites = total # of sites on a gene, G = # sites bound M = mRNA, Po = unmodified protein, Pt = modified protein V = Volume Transcription trans*V or 0 +M Translation tl*M +Po Protein Modification conv*Po -Po, +Pt M degradation degM*M -M Po degradationdegPo*Po -Po Pt degradationdegPt*Pt -Pt Binding to DNAbin/V(numsites - G)*Pt -Pt, +G Unbinding to DNAunbin*G -G
6
How would you simulate this? Choose which reaction happens next –Find next reaction –Update species by stochiometry of next reaction –Find time to this next reaction
7
How to find the next reaction Choose randomly based on their reaction rates trans*Vtl*M conv*Po degM*MdegPo*PodegPt*Pt bin/V(numsites - G)*Pt unbin*G Random #
8
Now that we know the next reaction modifies the protein Po = Po - 1 Pt = Pt + 1 How much time has elapsed –a 0 = sum of reaction rates –r 0 = random # between 0 and 1
9
This method goes by many names Computational Biologists typically call this the Gillespie Method –Gillespie also has another method Material Scientists typically call this Kinetic Monte Carlo
10
Myth 1: “Mass Action Formulations do not account for Stochasticity”
11
Consider a simple model inspired by the circadian clock in Cyanobacteria AB C
12
Here a protein can be in 3 states, A, B or C We start the system with 100 molecules of A Assume all rates are 1, and that reactions occur without randomness (it takes one time unit to go from A to B, etc.) AB C
14
Mass Action Representation
16
Matlab simulation
17
Mass Action represents a limiting case of Stochastics Mass action and stochastic simulations should agree when certain “limits” are obtained Mass action typically represents the expected concentrations of chemical species (more later)
18
Myth 2: Stochastic and Mass Action Approaches agree only if there are enough molecules
20
What matters is the number of reactions This is particularly important for reversible reactions By the central limit theorem, fluctuations dissapear like n -1/2 There are almost always a very limited number of genes, –Ok if fast binding and unbinding
21
There are several representations in between Mass Action and Gillespie Chemical Langevin Equations Master Equations Fokker-Planck Moment descriptions
22
We will illustrate this with an example Kepler and Elston Biophysical Journal 81:3116
23
Master Equations describe how the probability of being in each state
24
Sometimes we can solve for the mean and variance
25
Distribution of molecules often looks Gaussian
26
Moment Descriptions Gaussian Random Variables are fully characterized by their mean and standard deviation We can write down odes for the mean and standard deviation of each variable However, for bimolecular reactions, we need to know the correlations between variables (potentially N 2 )
27
Towards Fokker Planck Let’s divide the master equation by the mean m *. Although this equation described many states, we can smooth the states to make a probability distribution function
28
Note If 1/m* is small, we can then derive a simplifed Version of the Master equations
30
Chemical Langevin Equations If we don’t want the whole probability distribution, we can sometimes derive a stochastic differential equation to generate a sample
31
Adalsteinsson et al. BMC Bioinformatics 5:24
32
Examples Transcription Control Lac Operon Oscillations Accounting for diffusion
33
Rossi et al. Molecular Cell
36
Ozbudak et al. Nature 427:737
38
Guantes and Poyatos PLoS Computational Biology 2:e30
40
SNIC Bifurcation Invariant Circle Limit Cycle x2x2 p1p1 node saddle Saddle-Node on an Invariant Circle max min max SNIC
41
Hopf Bifurcation x2x2 p1p1 stable limit cycle sss uss slc max min
42
Noise Induced oscillations
43
Liu et al. Cell 129:605
44
3-D Gillespie http://www.math.utah.edu/~isaacson/3dmodel.html
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.