---
license: mit
configs:
  - config_name: default
    data_files: ["data/retina_dataset.csv"]
---

## ⚗️ ReTiNA: A Benchmark Dataset for LC-MS Retention Time Modeling

ReTiNA is a large open-source dataset for training machine learning models to predict small molecule retention times in LC-MS workflows.

Available models trained on the ReTiNA dataset are available at this [Hugging Face Repository](https://huggingface.co/natelgrw/ReTiNA-Models). 

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

This dataset is actively expanding with new experimental retention time values from the Coley Research Group at MIT, ensuring it remains a growing resource for optical property prediction.

ReTiNA is designed for use in:

- Estimating retention times for new compound–environment combinations
- Aiding in peak assignment during LC-MS method development
- Training ML models for retention time prediction under specific conditions

## 📈 The ReTiNA Dataset

The ReTiNA dataset contains:

- 119,039 unique molecule–environment combinations, the largest singular LC-MS retention time dataset of its kind to date
- Experimentally measured retention times, in seconds, curated from public datasets, benchmark papers, and literature
- Chemical descriptors for 105,809 unique compounds, 6 unique solvents, and 8 unique additives

73 distinct LC-MS setup environments are used in ReTiNA. Each environment consists of:

- Solvent mixtures A and B, consisting of solvents and solvent additives contributing to pH
- The mobile phase gradient used, defined by the percentage of solvent mixture B over time (min)
- The chromatographic technique used - either HILIC or reverse phase 
- LC-MS column dimensions, in terms of column length (mm), internal diameter (mm), and particle size (µm)
- The mobile phase flow rate, measured in mL/min
- The column temperature, measured in degrees Celsius

The ReTiNA dataset is divided into different scaffold, cluster, and method splits, which can be accessed in the `data` directory and are using in model evaluation.

## 📋 Data Sources Used

Detailed information on the data sources comprising ReTiNA can be found in the data folder.

## ✒️ Citation

If you use this code in a project, please cite the following:

```
@dataset{retinadataset,
  title={ReTiNA: A Benchmark Dataset for LC-MS Retention Time Modeling},
  author={Leung, Nathan},
  institution={Coley Research Group @ MIT}
  year={2025},
  howpublished={\url{https://huggingface.co/datasets/natelgrw/ReTiNA}}
}
```