mjohn218/KineticAssembly_AD

Python code for numerical simulations of self-assemly with optimization by automatic differentiation

KineticAssembly_AD

Python code for numerical simulations of self-assemly with optimization by automatic differentiation

Installation

The easiest way to install the simulator is to clone this repo and then build an environment containing all dependencies using the provided base_requirements.txt file. In order to do this, you will need to have an up to date version of the Anaconda package manager.

• First, clone this repository into the desired directory on your system with git clone https://github.com/mjohn218/KineticAssembly_AD.git.
• Enter the directory KineticAssembly_AD and run conda create --name <env> --file base_requirements.txt, where <env> is the desired name of your new environment. NOTE: This requirements file only includes dependencies available from conda or pip. For any application involving Rosetta, e.g. estimating free energies from PDB structures, you will need to also install PyRosetta in the environment.
• Run conda activate <env> to activate the new environment.

Documentation

Detailed functionality and documentation can be found in the Jupyter Notebooks located in the docs directory. A UserGuide is provided in the docs folder which desribes the requirements of an input file. Further detailed instructions on creating Reaction Networks, running simulations and optimization can be found within jupyter notebooks in the 'results' folder. You can start the Jupyter server by activating the conda environment and running the command jupyter notebook. This should open a browser window showing the current directory. You can then open the docs folder and then any of the notebooks therewithin.

The module consists of the following components:

• The ReactionNetwork class provides methods for generating a full set of possible system states and interactions from a list of pairwise rule and free energies specified in a input .pwr file.
• The VectorizedReactionNetwork class takes a ReactionNetwork object as input and converts its NetworkX explicit graph representation into a PyTorch tensor representation. Also provides methods for some computations on the system.
• The VecSim class, in which a vectorized deterministic simulator runs a rule based simulation on the system specified by a VectorizedReactionNetwork.
• The Optimizer class runs a certain number of simulations and adjusts input parameters in order to optimize a metric of complex yield.
• The EqSolver class takes a ReactionNetwork object as input and finds the equilibrium solution using a numerical solver.
• The EnergyExplorer class is somewhat separate. It takes as input a ReactionNetwork object and and a directory of .pdb files, one for each subunit monomer, and generates a list of approximate free energies for each of the pairwise reactions. Requires PyRosetta.

Results

The 'results' folder contains jupyter notebooks for different types of optimization protocols as outlined in the paper.