Cloud-COPASI supports a variety of simulation and analysis task types. These tasks have the common property that they can be split into a number of smaller independent jobs to be run in parallel, and therefore have the potential to significantly reduce the run time.
Models should be prepared and saved in the COPASI software before being uploaded to Cloud-COPASI. All recent versions of COPASI are supported, though we recommend using the most recent version.
Each of the task types have different requirements for how models should be prepared. Refer to the reference below for full details.
The following task types are supported:
Cloud-COPASI is able to automate and parallelize the global sensitivity analysis procedure, as described in the paper A new strategy for assessing sensitivities in biochemical models.
All parameters that you wish to vary should be added to the Optimization task, with initial values appropriate upper and lower bounds set. An appropriate optimization method, such as 'Particle Swarm' should also be set. In addition, the parameter you wish to calculate the sensitivities against (for example, the flux through a reaction, or the concentration of a metabolite) should be set as the Function the Sensitivities task as a Single Object.
Cloud-COPASI will split this task up into two parallel jobs per parameter specified in the Optimization task -- one job to minimize the sensitivity and one job to maximise it.
Cloud-COPASI produces a tab-separated text file containing, for each parameter in the Optimization task, the maximum and minimum values found for the parameter specified in the Sensitivities task. In addition, information is returned on the amount of CPU time taken per optimization run, and the number of function evaluations taken by the optimization algorithm.
Download an example model that has been configured to run the Sensitivity Optimization task.
Cloud-COPASI provides an easy to use environment for running multiple stochastic simulations. The simulations are automatically split into parallel jobs and run on the compute pool. Once completed, the results are automatically collated, and particle number means and standard deviations calculated.
The model should be prepared as if a single Time Course task were to run. A stochastic or hybrid algorithm should be selected, and all other parameters for the Time Course task set as desired.
Cloud-COPASI will automatically split this task into a number of sub-jobs, each of which can be executed in parallel on the compute pool. The number of repeats performed for each parallel job will depend on the time taken to simulate a single run of the Time Course task. Cloud-COPASI aims to make each Stochastic Simulation job submitted to the compute pool run for a constant length of time. If a single stochastic run of the Time Course task takes longer than this time, then a single stochastic run will be performed for each parallel job.
Typically, the running time of each job is set to be approximately equal to 20 minutes, but can be adjusted by the System Administrator. Note that, due to differences in the processing power of different machines in the compute pool, the running time of individual jobs may vary considerably.
Cloud-COPASI will automatically calculate particle number means and standard deviations for all species in the model at each time point. These are available to download as a tab-separated text file by clicking 'Download the results', and can be plotted by clicking 'View the results'.
The output produced for each individual Time Course repeat is also available in the file raw_results.txt. This file is available by clicking 'Download results directory'.
Download an example model configured to run the Stochastic Simulation task.
Cloud-COPASI will take a Parameter Scan set up in COPASI and automatically split it into smaller chunks to be run in parallel. In cases where multiple scan tasks/repeats are nested, only the the top-level scan will be split. The smaller chunks are then submitted to the compute pool, and the results collated as if the scan task had been run on a single machine.
The Parameter Scan task in COPASI should be set up as though the scan were to take place on the local machine. Unlike the other Cloud-COPASI tasks, Parallel Scan requires that a report be set for the Parameter Scan task. This report must contain output you wish to generate.
Like the Stochastic Simulation task, Cloud-COPASI aims to split the Scan task into smaller jobs that can be run in parallel on the compute pool, aiming to make each smaller job run for a constant length of time. If nested scans or repeats are set, only the top-level scan or repeat can be split. This should be taken into account when deciding on the order of the nested items.
Output is generated according to the report set for the Parameter Scan task. You must set this up manually using COPASI. The output is collated as though the Parameter Scan task had been run on a single machine.
Download an example model configured to run the Parallel Scan task.
This feature runs the optimization task a set number of times, splitting into multiple parallel jobs where necessary. The best value from the multiple number of runs is extracted, though the results of every optimization run are available for download too if necessary.
The Optimization task should be set up as though a single optimization was to take place on the local machine. All parameters should be set as necessary.
Like the Stochastic Simulation task, Cloud-COPASI aims to split the Optimization Repeat task into a number of small jobs, each of which will be executed in parallel on the compute pool. The number of repeats per job depends on the time taken to perform a single repeat.
Cloud-COPASI will automatically create a report for each Optimization repeat containing the best optimization value and the values of all variable parameters. Cloud-COPASI will then search the output files to find the best optimization value and associated parameter values.
The output from each optimization repeat is available in the file raw_results.txt, which can be obtained by downloading the results directory.
Download an example model configured to run the Optimization Repeat task.
Similar to the optimization repeat task above, this feature runs the parameter estimation task multiple times, splitting into parallel jobs where necessary. The best set of parameters values are then selected, though the results of each parameter estimation run are available to download if necessary.
> The Parameter Estimation task should be set up as though a single parameter estimation was to take place on the local machine. All parameters should be set as necessary, and any experimental data imported. Note that, when importing experimental data, all data files should be located in the same directory on the local machine as the model file.
Like the Stochastic Simulation task, Cloud-COPASI aims to split the Parameter Estimation Repeat task into a number of small jobs, each of which will be executed in parallel on the compute pool. The number of repeats per job depends on the time taken to perform a single repeat.
Cloud-COPASI will automatically create a report for each Parameter Estimation repeat containing the best objective value value and the values of all variable parameters. After all repeats have finished, Cloud-COPASI will search the output files to find the best objective value and associated parameter values.
Alternatively, by checking the appropriate box when submitting the task, a custom report can be used. This must be created manually using COPASI, and set for the Parameter Estimation task. Cloud-COPASI will try to process the output from any custom report; for this to succeed, the following fields must be placed (in order) at the end of the report:
TaskList[ParameterEstimation].(Problem)Parameter Estimation.Best Parameters
TaskList[ParameterEstimation].(Problem)Parameter Estimation.Best Value
TaskList[ParameterEstimation].(Problem)Parameter Estimation.(Timer)CPU Time
TaskList[ParameterEstimation].(Problem)Parameter Estimation.Function Evaluations
The output from each optimization repeat is available in the file raw_results.txt, which can be obtained by downloading the results directory.
Download an example model configured to run the Parameter Estimation task here, and download the associated data file here.
This feature runs the optimization task using different algorithms. Each algorithm runs as a separate parallel job. Cloud-COPASI will automatically determine which algorithm(s) found the best result.
The Optimization task should be configured as though a single optimization was to take place on the local machine. All parameters should be set as desired, except for those relating to the optimization algorithm, which will be set when submitting the model to Cloud-COPASI.
Cloud-COPASI creates a separate job to run on the compute pool for each optimization algorithm.
Cloud-COPASI will automatically create a report for the Optimization task containing the best optimization value, along with any associated variable parameter values.
After each optimization algorithm has run on the compute pool, Cloud-COPASI will go through the output and find the best value(s) and associated variable parameter values, and will list these alongside the name of the algorithm(s) which found the result.
The output for each optimization algorithm is available by downloading the results directory. Each output is named according to the name of the algorithm, e.g. particle_swarm_out.txt
Download an example model configured to run the Multiple Optimizations with Different Algorithms task.
The raw mode task type differs from the other tasks in that it allows any COPASI task to be run with a specified number of repeats. For each task that you wish to run, simply check the box marked 'executable' in the top-right of the COPASI interface when you are preparing the model. For each task that is run, you must create and set an output report for that task.
Raw mode allows you to specify custom arguments for the COPASI executable. These must include the variable '$filename' to represent the input file, and '--save $new_filename' to save the updated model.
Raw mode is not able to use the load balancing algorithm - each repeat is run as a separate parallel job.
A load balancing algorithm is used to ensure that tasks submitted into an optimum number of parallel jobs. See the publication Condor-COPASI: high-throughput computing for biochemical networks for full details on how this works. If you wish, you can bypass the load balancing algorithm. Doing this will mean one parallel job will be submitted for each task repeat. Be very careful when selecting this option, since it has the potential to overload the system with very large numbers of parallel jobs!