This is the official codebase for AGILE Platform: A Deep Learning-Powered Approach to Accelerate LNP Development for mRNA Delivery.
🥳 Updates: AGILE has been accepted to Nature Communications!
AGILE (AI-Guided Ionizable Lipid Engineering) platform streamlines the iterative development of ionizable lipids, crucial components for LNP-mediated mRNA delivery. This platform brings forth three significant features:
🧪 Efficient design and synthesis of combinatorial lipid libraries
🧠 Comprehensive in silico lipid screening employing deep neural networks
🧬 Adaptability to diverse cell lines
It also significantly truncates the timeline for new ionizable lipid development, reducing it from potential months or even years to weeks ⏱️!
An overview of AGILE can be seen below:
Clone the github repo and set up conda environment
# Clone the GitHub Repository
$ git clone <this-repo-url>
# Create a new environment
$ conda create --name agile python=3.9 -y
$ conda activate agile
# Install PyTorch and torchvision with CUDA support. Make sure the versions are compatible with your CUDA version.
$ pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
$ pip install torch-geometric==2.2.0 torch-sparse==0.6.16 torch-scatter==2.1.0 -f https://data.pyg.org/whl/torch-1.12.0+cu113.html
$ pip install -r requirements.txt
We have provided the fine-tuning library and candidate library used in the paper in data.zip
, extract the zip file under ./data
folder.
The pre-trained AGILE model on the 60k virtual lipid library can be found in ckpt/pretrained_agile_60k
. If your data significantly differs from the 60k virtual lipid library, the pre-trained AGILE model might not perform optimally. In such cases, you can pre-train the model with your own dataset to potentially achieve better results.
Steps to Pre-train with Your Data:
-
Obtain Pre-trained Base Models: Download the pre-trained MolCLR models, which serve as a starting point for further training. These models are available at here.
-
Set Up the Model Directory: Place the downloaded MolCLR model files in the
./ckpt
directory within your project folder. This ensures they are properly accessed by the training script. -
Configure Training Settings: Open the
config_pretrain.yaml
file and make the following adjustments:
load_model
: Change this to the model name of your downloaded MolCLR model.data_path
: Specify the path to your dataset where the training data is stored.
$ python pretrain.py config_pretrain.yaml
To fine-tune the AGILE pre-trained model for ionizable lipid prediction on the specific cell lines, you can modify the configurations in config_finetune.yaml
.
If you would like to fine-tune AGILE with your own dataset, create your own task_name
in the config file, and modify the following fields in the finetune.py
:
config["dataset"]["task"] = "regression" # keep it the same
config["dataset"]["data_path"] = "data/finetuning_set_smiles_plus_features.csv" # change it to the path of your own fine-tunning dataset
target_list = ["expt_Hela"] # change it to the column name of the regression labels
config["dataset"]["feature_cols"] = get_desc_cols(config["dataset"]["data_path"]) # keep it the same if you have additional features
config["model"]["pred_additional_feat_dim"] = len(config["dataset"]["feature_cols"]) # keep it the same if you have additional features
Then run:
$ python finetune.py config_finetune.yaml
The fine-tuned AGILE model will be stored in ./finetune
.
To perform model inference with the fine-tuned AGILE model, you can run the following command:
$ python infer_vis.py <folder name of the fine-tuned model>
Note that the 'infer_vis.py' will pick up the config yaml file from the fine-tuned AGILE model folder. So the above command will perform model inference with the specified AGILE fine-tuned model on the fine-tuning dataset. To perform inference on new data, you will need to modify the config file with a new task_name
and modify the data_path
field in the infer_vis.py
:
config["dataset"]["task"] = "regression" # keep it the same
config["dataset"]["data_path"] = "data/candidate_set_smiles_plus_features.csv" # change it to the path of your own inference dataset
target_list = ["desc_ABC/10"] # it will be the dummy label for visualization
config["dataset"]["feature_cols"] = get_desc_cols(config["dataset"]["data_path"]) # keep it the same if you have additional features
config["model"]["pred_additional_feat_dim"] = len(config["dataset"]["feature_cols"]) # keep it the same if you have additional features
The predicted output (in .csv file) and visualization plots (in .png files) will be stored in the same fine-tuned AGILE model folder.
@article{xu2023agile,
title={AGILE Platform: A Deep Learning-Powered Approach to Accelerate LNP Development for mRNA Delivery},
author={Xu, Yue and Ma, Shihao and Cui, Haotian and Chen, Jingan and Xu, Shufen and Wang, Kevin and Varley, Andrew and Lu, Rick Xing Ze and Bo, Wang and Li, Bowen},
journal={bioRxiv},
pages={2023--06},
year={2023},
publisher={Cold Spring Harbor Laboratory}
}