ChemTSv2 is a refined and extended version of ChemTS1 and MPChemTS2. The original implementations are available at https://github.com/tsudalab/ChemTS and https://github.com/yoshizoe/mp-chemts, respectively.
ChemTSv2 provides:
- easy-to-run interface by using only a configuration file
- easy-to-define framework for users' any reward function, molecular filter, and tree policy
- various usage examples in the GitHub repository
Click to show/hide requirements
- python: 3.7
- rdkit: 2021.03.5
- tensorflow: 2.5.0
- pyyaml
- pandas
- joblib
- mpi4py: 3.0.3 (for massive parallel mode)
Click to show/hide the instruction
cd YOUR_WORKSPACE
python3.7 -m venv .venv
source .venv/bin/activate
pip install --upgrade chemtsv2
Click to show/hide the instruction
NOTE: You need to run ChemTSv2-MP on a server where OpenMPI or MPICH is installed. If you can't find `mpiexec` command, please consult your server administrator to install such an MPI library.If you can use/prepare a server with MPI environment, please follow the (a) instruction; otherwise, please follow the (b) instruction.
cd YOUR_WORKSPACE
python3.7 -m venv .venv
source .venv/bin/activate
pip install --upgrade chemtsv2
pip install mpi4py==3.0.3
conda create -n mpchem python=3.7
# swith to the `mpchem` environment
conda install -c conda-forge openmpi cxx-compiler mpi mpi4py=3.0.3
pip install --upgrade chemtsv2
git clone [email protected]:molecule-generator-collection/ChemTSv2.git
cd ChemTSv2
Please refer to reward/README.md
.
An example of reward definition for LogP maximization task is as follows.
from rdkit.Chem import Descriptors
import numpy as np
from reward.reward import Reward
class LogP_reward(Reward):
def get_objective_functions(conf):
def LogP(mol):
return Descriptors.MolLogP(mol)
return [LogP]
def calc_reward_from_objective_values(objective_values, conf):
logp = objective_values[0]
return np.tanh(logp/10)
The explanation of options are described in the Support option/function section.
The prepared reward file needs to be specified in reward_setting
.
For details, please refer to a sample file (config/setting.yaml).
If you want to pass any value to calc_reward_from_objective_values
(e.g., weights for each value), add it in the config file.
chemtsv2 -c config/setting.yaml
mpiexec -n 4 chemtsv2-mp --config config/setting_mp.yaml
Target | Reward | Config | Additional requirement | Ref. |
---|---|---|---|---|
LogP | logP_reward.py | setting.yaml | - | - |
Jscore | Jscore_reward.py | setting_jscore.yaml | - | 1 |
Absorption wavelength | chro_reward.py | setting_chro.yaml | Gaussian 163 via QCforever4 |
5 |
Upper-absorption & fluorescence wavelength |
fluor_reward.py | setting_fluor.yaml | Gaussian 163 via QCforever4 |
6 |
Kinase inhibitory activities | dscore_reward.py | setting_dscore.yaml | LightGBM7 | 8 |
Docking score | Vina_binary_reward.py | setting_vina_binary.yaml | AutoDock Vina9 | 10 |
Option | Single process | Massive parallel | Description |
---|---|---|---|
c_val |
β | β | Exploration parameter to balance the trade-off between exploration and exploitation. A larger value (e.g., 1.0) prioritizes exploration, and a smaller value (e.g., 0.1) prioritizes exploitation. |
threshold_type |
β | βοΈ | Threshold type to select how long (hours ) or how many (generation_num ) molecule generation to perform. Massive parallel mode currently supports only the how long (hours ) option. |
hours |
β | β | Time for molecule generation in hours |
generation_num |
β | β¬ | Number of molecules to be generated. Please note that the specified number is usually exceeded. |
expansion_threshold |
β | β¬ | (Advanced) Expansion threshold of the cumulative probability. The default is set to 0.995. |
simulation_num |
β | β¬ | (Advanced) Number of rollout runs in one cycle of MCTS. The default is set to 3. |
flush_threshold |
β | β¬ | Threshold for saving the progress of a molecule generation. If the number of generated molecules exceeds the threshold value, the result is saved. The default is set to -1, and this represents no progress is to be saved. |
Molecule filter | β | β | Molecule filter to skip reward calculation of unfavorable generated molecules. Please refer to filter/README.md for details. |
RNN model replacement | β | β | Users can switch RNN models used in expansion and rollout steps of ChemTSv2. The model needs to be trained using Tensorflow. model_json specifies the JSON file that contains the architecture of the RNN model, and model_weight specifies the file in H5 format that contains a set of the values of the weights. |
Reward replacement | β | β | Users can use any reward function as long as they follow the reward base class (reward/reward.py). Please refer to reward/README.md for details. |
Policy replacement | β | β¬ | (Advanced) Users can use any policy function as long as they follow the policy base class (policy/policy.py). Please refer to policy/README.md for details. |
Restart | π° | β¬ | Users can save the checkpoint file and restart from the file. If users want to save a checkpoint file, set save_checkpoint to True and specify the file name in checkpoint_file . If users want to restart from the checkpoint, set restart to True and specify the checkpoint file in checkpoint_file . |
- β indicates that the option/function is supported.
- βοΈ indicates that the option/function is partially supported.
- π° indicates that the option/function is beta version.
- β¬ indicates that the option/function is NOT supported.
Filter functions are described in filter/README.md.
You can extend the SMILES string you input.
In this case, you need to put the atom you want to extend at the end of the string and run ChemTS with --input_smiles
argument as follows.
chemtsv2 -c config/setting.yaml --input_smiles 'C1=C(C)N=CC(N)=C1C'
If you want to use GPU, run ChemTS with --gpu GPU_ID
argument as follows.
chemtsv2 -c config/setting.yaml --gpu 0
This package is distributed under the MIT License.
- Shoichi Ishida ([email protected])
- Kei Terayama ([email protected]).
Footnotes
-
Yang, X., Zhang, J., Yoshizoe, K., Terayama, K., & Tsuda, K. (2017). ChemTS: an efficient python library for de novo molecular generation. Science and Technology of Advanced Materials, 18(1), 972β976. https://doi.org/10.1080/14686996.2017.1401424 β© β©2
-
Yang, X., Aasawat, T., & Yoshizoe, K. (2021). Practical Massively Parallel Monte-Carlo Tree Search Applied to Molecular Design. In International Conference on Learning Representations. https://openreview.net/forum?id=6k7VdojAIK β©
-
Frisch, M. J. et al. Gaussian 16 Revision C.01. 2016; Gaussian Inc. Wallingford CT. β© β©2
-
Sumita, M., Terayama, K., Tamura, R., & Tsuda, K. (2022). QCforever: A Quantum Chemistry Wrapper for Everyone to Use in Black-Box Optimization. Journal of Chemical Information and Modeling, 62(18), 4427β4434. https://doi.org/10.1021/acs.jcim.2c00812 β© β©2
-
Sumita, M., Yang, X., Ishihara, S., Tamura, R., & Tsuda, K. (2018). Hunting for Organic Molecules with Artificial Intelligence: Molecules Optimized for Desired Excitation Energies. ACS Central Science, 4(9), 1126β1133. https://doi.org/10.1021/acscentsci.8b00213 β©
-
Sumita, M., Terayama, K., Suzuki, N., Ishihara, S., Tamura, R., Chahal, M. K., Payne, D. T., Yoshizoe, K., & Tsuda, K. (2022). De novo creation of a naked eyeβdetectable fluorescent molecule based on quantum chemical computation and machine learning. Science Advances, 8(10). https://doi.org/10.1126/sciadv.abj3906 β©
-
Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., β¦ Liu, T.-Y. (2017). Lightgbm: A highly efficient gradient boosting decision tree. Advances in Neural Information Processing Systems, 30, 3146β3154. β©
-
Yoshizawa, T., Ishida, S., Sato, T., Ohta, M., Honma, T., & Terayama, K. (2022). Selective Inhibitor Design for Kinase Homologs Using Multiobjective Monte Carlo Tree Search. Journal of Chemical Information and Modeling, 62(22), 5351β5360. https://doi.org/10.1021/acs.jcim.2c00787 β©
-
Eberhardt, J., Santos-Martins, D., Tillack, A. F., & Forli, S. (2021). AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings. Journal of Chemical Information and Modeling, 61(8), 3891β3898. https://doi.org/10.1021/acs.jcim.1c00203 β©
-
Ma, B., Terayama, K., Matsumoto, S., Isaka, Y., Sasakura, Y., Iwata, H., Araki, M., & Okuno, Y. (2021). Structure-Based de Novo Molecular Generator Combined with Artificial Intelligence and Docking Simulations. Journal of Chemical Information and Modeling, 61(7), 3304β3313. https://doi.org/10.1021/acs.jcim.1c00679 β©