This package, pyODEM (python Observable-driven Ddesign-of Effective Molecular-models), adjusts model parameters based upon experimental data. The method was developed in the Clementi Group at Rice University.
#Prerequisites
The TensorDuck/langevin_model repository needs to be cloned and in your PYTHONPATH variable for using the model_loaders.Langevin() loader.
The ajkluber/model_builder repository needs to be cloned and in your PYTHONPATH variable for using the model_loaders.Proteins() loader.
Main packages and methods to be aware of for the end user is:
##model_loaders
Modules with methods for loading simulation data and analyzing it.
model_loaders.Langevin(): 1-D langevin dynamics data. See package TensorDuck/langevin_model.
model_loaders.Proteins(): Loading a protein topology and its associated potential functions. See package ajkluber/model_builder.
##observables
Modules for loading experimental results, computing Q values, and computing observables from simulation data.
observables.ExperimentalObservables.add_histogram(): Adds a histogram data and associated observables for 1-D position data.
##estimators
Modules for estimating the Maximum Likelihood set of parameters for a model based on some experimental data (obsevables).
estimators.max_likelihood_estimate(): Estimates the most likely set of model parameters given the data.
Currently supports several different solvers. Options are: simplex, cg, anneal and custom. See documentation.
##Examples
example_0: Compute the Q-value for the data files present in the folder.
example_1: Compute a new set of model parameters for the Langevin model using the data files present in the folder.
To Run, use execute multirun.sh in the folder to run a langevin 1-D simulation using the current set of starting parameters. There are 3-steps per an iteration in multirun.sh:
python -m langevin_model.simulation sim --name $name --steps 1000000: Executes a langevin 1-d simulation. You can change the number of steps it takes. More steps means more data to analyze so it will take longer to simulate and analyze, with ~O(N) scaling. The time function is there for diagnostic purposes. It will generate a new folder iteration_%d where %d is the iteration number.
python -m run: Executes the analysis script in the folder called run.py. This will analyze using the max_likelihood method and output a new set of parameteres into iteration_%d/newton/params
python -m langevin_model.simulation next --name $name --start: Update the input files for the next simulation run (i.e. update model parameters).
example_2: Compute a new set of model parameters on a Protein model. To run, use the jupyter notebook file.