mdlearn

Machine learning of the thermodynamic properties of molecular liquids

This is the enhanced version of our previously published work
Predicting Thermodynamic Properties of Alkanes by High-throughput Force Field Simulation and Machine Learning
https://doi.org/10.1021/acs.jcim.8b00407

Python Packages

• Create ml environment
  conda create -n ml

• Activate ml environment
  source activate ml

• Add tsinghua pytorch mirror conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
conda config --set show_channel_urls yes

• Install packages
  conda install matplotlib scikit-learn pytorch
conda install -c openbabel openbabel
conda install -c rdkit rdkit

Steps

Following is an example of learning critical temperature of hydrocarbon using two simple fingerprints

These scripts are located at directory run

• Activate ml environment

source activate ml

• Calculate fingerprints

./gen-fp.py -i ../data/nist-CH-tc.txt -e morgan1-200,topological-500,simple -o fp

• Split data to train/validate datasets using 5-Fold cross-validation

./split-data.py -i ../data/nist-CH-tc.txt -o fp

• Train the model

./train.py -i ../data/nist-CH-tc.txt -t tc -f fp/fp_morgan1-200,fp/fp_simple -p fp/part-1.txt -o out

• Predict property for new molecules

./predict.py -d out -e predefinedmorgan1-200,simple -i CCCCCC

Note that if you train the model with morgan1 or morgan fingerprint, you should use predefinedmorgan1 or predefinedmorgan to predict

Useful scripts

These scripts are located at directory scripts

• Generate similar structures for those molecules which give depressing results (Currently only support hydrocarbon)

../scripts/gen-similar.py out/error-0.1.txt