From SMILES to Enhanced Molecular Property Prediction: A Unified Multimodal Framework with Predicted 3D Conformers and Contrastive Learning Techniques (JCIM 2024)

This repository contains the official implementation of "From SMILES to Enhanced Molecular Property Prediction: A Unified Multimodal Framework with Predicted 3D Conformers and Contrastive Learning Techniques", published in JCIM 2024.

We introduce a novel molecular property prediction framework that takes SMILES strings as input and extends it to a multimodal approach by incorporating predicted 3D conformer representations. This allows the model to capture both the sequential character structure of SMILES and the three-dimensional spatial structure of molecules.

Key highlights of the framework:

• Contrastive Learning: Utilizes InfoNCE loss to align SMILES and conformer embeddings.
• Task-Specific Losses: Employs ConR for regression tasks and SupCon for classification tasks.
• Feature Distribution Smoothing (FDS): Addresses data imbalance challenges commonly encountered in drug discovery.

Requirements

• Python 3.8.5

• Install required dependencies:

  pip install -r requirements.txt
  

• Uni-Core: Install the Uni-Core library from GitHub:

---

Pretrained Models

The pretrained Uni-Mol and ChemBERTa models used in this framework are hosted on HuggingFace:

• Download them from ndlong/mm-dti.

Clone the model weights into the weights folder:

git clone https://huggingface.co/ndlong/mm-dti weights

---

Dataset

We tested our framework on:

• SARS-CoV-2 Drug Docking Score Prediction
• MoleculeNet Datasets for molecular property prediction.
• Custom Kinase Inhibitor Datasets for JAK-1, JAK-2, and MAPK-14.

For access to refined datasets, please contact Dr. Binh Nguyen.

---

Usage

To reproduce results from the paper, configure hyperparameters in finetune.py and run the training script:

## Start training
python finetune.py

---

Acknowledgements

This repository builds upon the following projects:

• Uni-Mol
• ConR
• imbalanced-regression

We are grateful to the contributors of these projects for their valuable work!

---

Citation

If you find this work useful in your research, please cite:

@article{doi:10.1021/acs.jcim.4c01240,
  author = {Nguyen, Long D. and Nguyen, Quang H. and Trinh, Quang H. and Nguyen, Binh P.},
  title = {From SMILES to Enhanced Molecular Property Prediction: A Unified Multimodal Framework with Predicted 3D Conformers and Contrastive Learning Techniques},
  journal = {Journal of Chemical Information and Modeling},
  doi = {10.1021/acs.jcim.4c01240},
  note ={PMID: 39641280},
  URL = {https://doi.org/10.1021/acs.jcim.4c01240}
}