Random forest models for predicting the coefficient of friction and adhesion for systems of two contacting functionalized monolayers.

This work is associated with the manuscript "Examining Chemistry-Property Relationships in Lubricating Monolayer Films through Molecular Dynamics Screening" currently under review, authored by Andrew Z. Summers, Justin B. Gilmer, Christopher R. Iacovella, Peter T. Cummings, and Clare McCabe at Vanderbilt University.

Installation

Use of this model requires several Python packages to be installed, as well as data obtained from molecular dynamics screening. Most of the required packages are located in the req.txt file. It is recommended to use the Anaconda package manager to create a new environment, as the packages can be pulled from this file directly.

The recommended installation instructions are as follows:

Clone this repository

>> git clone https://github.com/PTC-CMC/random_forest_tg.git

Create a new Anaconda environment

NOTE: This environment is only meant for MacOS, as certain packages like appnope are MacOS specific

>> conda env create -f environment-macos.yml

>> conda activate screening35

>> pip install -r requirements-macos.txt

Download data from MD screening

git clone https://github.com/PTC-CMC/terminal_group_screening.git
git clone https://github.com/PTC-CMC/terminal_groups_mixed.git

Install atools-ml package

git clone https://github.com/PTC-CMC/atools_ml.git
cd atools_ml
pip install .
cd ..

Using the models

The random forest models can be regenerated in a few seconds. Thus, rather than providing these models in a form already generated (such as a serialized form like pickle), the script herein re-creates the models on the spot. The script rf.py is used to regenerate the models and generate predictions for user-specified terminal group chemistries. These can be changed by opening the file and altering the "SMILES1" and "SMILES2" variables. Further instructions can be found inside the rf.py file.