Re: Help with stochastic simulation

[juergen]

Hello Sebastián,

(exact) stochastic simulations calculate each and every "reactive event" in the system. Each of these single reactions, i.e. one set of substrates are converted to products, constitutes one iteration in algorithms, such as Gillespie's Direct Method or Gibson and Bruck's Next Reaction Method. Therefore, the computation time of these algorithms scales roughly with the number of particles in the system, as with more particles in the system more reactions take place per unit time. This means that particle number are important to keep in mind. The compartment in your model is set to one litre. This means your system reaches particle numbers up to 1e+17, which results in very many simulation iterations. This explains that your simulations are very very slow or you even reach the internal limit for iterations per simulation step.

You can simply solve this problem by setting the compartment volume to values, such as 1e-13, which will keep the concentrations the same but scales down the particle numbers in you system, making stochastic simulations feasible.

Hope this helps.

Best wishes,

Juergen

On Tuesday, February 13, 2024 at 10:25:52 PM UTC+1 Sebastián Espinel Ríos wrote:

Hello World!

I am new to stochastic simulations in COPASI.

I am simulating a relatively simple network of chemical reactions. The deterministic simulation runs super fast (as expected), but when it comes to the stochastic one,  I keep getting:
"CTrajectoryMethod (12): Internal step limit exceeded."

Increasing the Internal step limit just makes the model "loop" forever or simply does not solve the issue.

I have compared my simulation settings with other stochastic models in COPASI, such as the one in the documentation, and it seems like everything is fine. I cannot spot the issue.

I hereby attach the COPASI file just in case somebody wants to check on his end. I would appreciate your help a lot!

Best,

Sebastián

[juergen]

Yes, that is exactly right (on average).

On Wednesday, February 14, 2024 at 1:00:44 AM UTC+1 Sebastián Espinel Ríos wrote:

Thank you very much, Juergen! :)

Your suggestion indeed worked. So basically, the delta time of the stochastic method, e.g., in Gillespie's method, scales inversely with the total propensity, right? That is why, if I understand correctly, with large particle numbers I expect to have large propensities, then very small delta times, and thus very long computation times.