Computational Prediction of De-Esterification by Human Carboxylesterases 1 and 2

Author : Rüthemann Peter, André Fischer

Description

Computational Prediction of De-Esterification by Human Carboxylesterases 1 and 2 using Random Forests (RFs) trained on literature-derived dataset.

This repository contains all used structures as sdf or mae files, as well as the used data set for training. The predictions for the training, as well as the two test sets EXT_A and EXT_B are provided.

The two scripts Train.py and Prediction.py allow the reproduction of the results in table 1 and the generation of the Random Forest (RF) model.

Dataset:
    - input/datasets.csv:       Data sets with descriptor data and labels (y_true)

Random forest model
    - input/features.csv:       Selected features used for training
    - input/parameters.yaml:    Hyper parameters required for initialisation of sklearn Random Forest
    - model/RF_classifier.pkl:  Random forest model stored as pickle file

Results files:
    - prediction/TRAIN.csv:     Dataset with predictions for training set
    - prediction/EXT_A.csv:     Dataset with predictions for external test set A
    - prediction/EXT_B.csv:     Dataset with predictions for external test set B  

SetUp

1. Install conda
2. Install conda environment
   conda env create -f environment.yml

Training

1. Activate environment
   conda activate ester-prediction
2. Execute training of training set
   python3 Train.py

Prediction

1. Activate environment
   conda activate ester-prediction
2. Execute prediction of training and two test sets
   python3 Prediction.py