RedPred
RedPred: Redox Energy Prediction Tool for Redox Flow Battery Molecules
About RedPred Project:
-
RedPred is an redox energy prediction model for redox flow battery molecules that consists ensemble of 3 ML algorithms (Graph Conv Neural Nets, Deep Neural Nets, and Random Forest).
-
The model takes the SMILES notations of reactant molecules of the redox reaction as an input and predicts the redox reaction energy (Hartree).
-
RedPred is trained on RedDB [1] publicly available redox flow battery candidate molecules dataset.
-
The performance of the RedPred is 0.0036 and 0.0043 Hartree MAE on the test-1 and test-2 sets, respectively.
-
If you are using the predictions from RedPred on your work, please cite these papers: [1, 2]
-
[1] Sorkun, Elif, et al. (2021). RedDB, a computational database of electroactive molecules for aqueous redox flow batteries.
-
[2] In preparation (will be updated soon)
-
Workflow - IMAGE :
Project Files:
Data: Contains the row and processed data files
Preprocess: Contains data preprocessing including removing missing values and test/train splitting (requires different dependencies, to reproduce it please check the requirements on the folder)
Ensemble: Contains ensembling process of the selected 3 models.
Models: Contains final code file of 5 models that we used for RedPred project and ECFC encoder file.
Dependencies:
- python=3.7.9 (requires "conda install python=3.7.9")
- rdkit=2020.09.1.0 (requires "conda install -c conda-forge rdkit=2020.09.1")
- scikit-learn=0.22.1
- deepchem==2.4.0
- numpy==1.18.5
- pandas==1.1.3
- tensorflow==2.3.2
- keras==2.4.3
- lightgbm==2.3.1
- xgboost==1.4.2
- h5py==2.10.0
Web application:
You can use the RedPred web application by following this link.
Report an Issue:
You are welcome to report a bug or contribuite to the RedPred project by filing an issue.
References:
[1]:
Developers:
-
Murat Cihan Sorkun :
-
Cihan Yatbaz :
-
Elham Nour Ghassemi :
This project developed at AMD LAB : https://www.amdlab.nl/