Lipid Nanoparticle Design with Composite Material Transformer (COMET)

===================================================================

• Python 3.10
• CUDA 11.6
• PyTorch 1.13.1
• Anaconda 23.1.0

1. Load the required modules and create a new Anaconda environment:

   conda create -n comet_env python=3.10
   source activate comet_env

2. Install dependencies:

   conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia
   pip install lmdb==1.4.0 ml-collections==0.1.1 numpy==1.23.4 scipy==1.9.3 tensorboardX==2.5.1 tqdm==4.64.1 tokenizers==0.13.2 pyprojroot==0.2.0 pandas==1.5.2 scikit-learn==1.2.0 rdkit-pypi==2022.9.3

3. Install Uni-Core, compatible with the specified versions:

   pip install https://github.com/dptech-corp/Uni-Core/releases/download/0.0.2/unicore-0.0.1+cu116torch1.13.1-cp310-cp310-linux_x86_64.whl

Preprocessing is done to make lmdb dataset (stored in processed_data_dirs/) from json files (stored in data_json/). Scripts for generating processed datasets from JSON files:

• experiments/preprocess_data_LANCE.ipynb: Processes data for LNPs' efficacy on DC2.4 and B16-F10 cells.
• experiments/preprocess_data_CACO2.ipynb: Includes CACO2 cell transfection data.
• experiments/preprocess_data_stability.ipynb: Processes data for lyophilized LNPs.

Training scripts for different models:

• Lipid-only LNP: experiments/training_script_LANCE_lipid_T01.py
• PBAE LNP: experiments/training_script_LANCE_PBAE_T02.py
• Different cell types (CACO2): experiments/training_script_caco2_T03.py
• Lyophilized LNP: experiments/training_script_stability_T04.py

Run inference using pretrained models:

• Lipid-only LNP: experiments/inference_script_LANCE_lipid_I01.py
• PBAE LNP: experiments/inference_script_LANCE_PBAE_I02.py

Predictions are stored as .pkl files in the specified output directories. These files end with ".out.pkl" in their names. Inference results will be outputted at location: dataset_output_dir>model_name>output_filename

• experiments/: Contains training and preprocessing scripts.
• experiments/task_schemas/: stores task schema files: dictionary file to specify key information of multiple datasets. The keys are names of the datasets, value is a dictionary containing tasks_schema_path, component_types_schema_path and np_prop_schema_path as keys
• experiments/data_json/: stores raw LNP data
• experiments/processed_data_dirs/: stores processed lmdb datasets for training and inference
• experiments/weights/: stores pretrained COMET weights, accompanying example inference scripts to use them are inference_script_*.py (download example weights here)
• unimol/: Core logic for model operations.
• ckp/: Pretrained model weights (download here)

Available pretrained weights for different models are listed under experiments/weights/. These can be used directly for deploying models and running inference scripts.

Code has been test on Linux and installation is expected to take less than 2 hours on a typical computer with internet connection.