---
license: mit
tags:
- chemistry
- molecular-property-prediction
- drug-discovery
datasets:
- natelgrw/ReTiNA
pipeline_tag: tabular-regression
---

# ReTiNA Models: Molecular Retention Time Prediction

A collection of machine learning models for predicting the retention time of chemical compounds in various LC-MS. These models use molecular descriptors and method encodings to predict chemical retention times, useful for automated compound identification.

Source code for the ReTiNA model collection is available at this [Github Repository](https://github.com/natelgrw/retina_models).

The ReTiNA dataset is available at this [Hugging Face Repository](https://huggingface.co/datasets/natelgrw/ReTiNA). 

## 🤖 Available Models

In retention time prediction, we recommend using **ReTiNA_XGB1**, as it has the highest overall prediction accuracy.

| Model | Architecture | Overall RMSE (s) | Overall MAE (s) | Overall R<sup>2</sup> |
|-----|-----|-----|-----|-----|
| **ReTiNA_XGB1** | XGBoost | 182.81 | 119.30 | 0.659 |
| **ReTiNA_MLP1** | PyTorch Residual MLP | 202.67 | 141.79 | 0.516 |

All models were evaluated across rigorous scaffold, cluster, and method splits.

## 📄 Citation

If you use a ReTiNA prediction model in your research, please cite:

```
@modelcollection{retinamodels,
  title={ReTiNA-Models: Machine Learning Models for LC-MS Retention Time Prediction},
  author={Leung, Nathan},
  institution={Coley Research Group @ MIT}
  year={2025},
  howpublished={\url{https://huggingface.co/natelgrw/ReTiNA-Models}},
}
```