This is a repository of the article "A modular artificial intelligence framework to facilitate fluorophore design".
This code has been tested on Linux(Ubuntu 22.04) and Windows(10,11), which only requires a standard computer with enough RAM to support the in-memory operations.
FLAME develop on python3.8 and mainly depends on these packages.
tqdm==4.66.2
torch==2.2.2
numpy==1.24.3
pandas==2.0.3
scikit-learn==1.3.2
packaging==24.0
typed-argument-parser==1.9.0
rdkit==2023.9.5
ase==3.22.1
tensorboardX==2.6.2.2
einops==0.7.0
h5py==3.10.0
padelpy==0.1.14
gdown==5.2.0
tables==3.8.0
pandas_flavor
We strongly recommand to install packages in conda environment.
After install conda, create an environmnet
conda create -n flame python==3.8
activate the environment and install the packages
conda activate flame
cd /dir/to/this/repo # !cd the dir you git clone or download this repository
pip install -r requirements.txt
and then you could use FLAME by python shell or jupyter.
We provide [UVVISML, ABTMPNN, Schnet, FCNN, GBRT and FLSF ] 6 models as baseline, and each model has example code in FLAME/scripts.
from FLAME import run_search
result_df = run_search(SMILES, similarity_limit, fingerprinter_type)
The basic format of trainning is:
from FLAME import flsf_train
flsf_train(model_path, train_data, valid_data, test_data, epoch)
And the prediction:
from FLAME import flsf_predict
flsf_predict(model_path, output_file, input_file=input_file)
or you can just run as a function:
from FLAME import flsf_predict
result = flsf_predict(model_path, smiles=SMILES)
We provided a FLAME_plugin for Reinvent to molecule design and molecuar optimization.
Fisrt download the FLAME_plugin.zip and unzip the file to /path/to/python/lib/site-packages/reinvent_plugins
You can Create reinvent toml config file with scoring stage like:
[stage.scoring]
type = "geometric_mean"
[[stage.scoring.component]]
[stage.scoring.component.flame_abs]
[[stage.scoring.component.flame_abs.endpoint]]
name = "flame" # user chosen name for output
weight = 10 # weight to fine-tune the relevance of this component
And flame_emi/flame_plqy/flame_e/stokes are also supported.
We'd love to hear from you. Please let us know if you have any comments or suggestions, or if you have questions about the code or the procedure. Also, if you'd like to contribute to this project please let us know. Please feel free to email us (yuchenzhu@zju.edu.cn or zyman@zju.edu.cn).
This code leveraged and built on top of Chemprop .
This project is covered under the MIT License.