README.md

---
pretty_name: Organic Solar Cell Molecule Property Prediction
license: other
task_categories:
- graph-ml
tags:
- chemistry
- organic-solar-cells
- molecular-property-prediction
- photovoltaics
- tabular
- graph-machine-learning
- ringformer
language:
- en
---

# Organic Solar Cell Molecule Property Prediction

<p align="left">
  <img src="assets/osc_examples.png" alt="OSC Examples" width="800">
</p>

## Summary
This Hugging Face dataset repository publishes **raw CSV tables** for **organic solar cell (OSC) molecule/device property prediction**, used in the paper:

*“RingFormer: A Ring-Enhanced Graph Transformer for Organic Solar Cell Property Prediction”* (AAAI 2025).

It includes **five datasets**: **CEPDB, HOPV, PFD, NFA, and PD**.

**Files in this repo**
- `CEPDB.csv`
- `HOPV.csv`
- `PFD.csv`
- `NFA.csv`
- `PD.csv`

**Important**
- This repository contains **raw/tabular CSV files only**.
- The **processed ring-graph data** and the code to generate it are provided in the RingFormer GitHub repository:  
  https://github.com/TommyDzh/RingFormer

---

## Dataset Overview

| DATASET | #GRAPHS | AVG. # NODES | AVG. # EDGES | AVG. # RINGS |
| ------: | ------: | -----------: | -----------: | -----------: |
| CEPDB   | 2.2M    | 27.6         | 33.3         | 6.7          |
| HOPV    | 350     | 42.7         | 49.3         | 7.5          |
| PFD     | 1055    | 77.1         | 84.2         | 8.2          |
| NFA     | 654     | 118.2        | 133.0        | 15.8         |
| PD      | 277     | 80.7         | 88.2         | 8.5          |

---

## Schema (Columns)

### `CEPDB.csv`
Header:
`smiles, PCE (%), Voc (V), Jsc, HOMO (eV), LUMO (eV)`

- `smiles`: molecule SMILES
- `PCE (%)`: power conversion efficiency (percent)
- `Voc (V)`: open-circuit voltage (volts)
- `Jsc`: short-circuit current density (unit as provided in the original file; commonly mA/cm²)
- `HOMO (eV)`: HOMO energy level (eV; typically negative)
- `LUMO (eV)`: LUMO energy level (eV; typically negative)

### `HOPV.csv`
Header:
`smiles,doi,inchlKEY,construction,architecture,complement,HOMO,LUMO,electrochemical_gap,optical_gap,PCE,V_OC,J_SC,fill_factor`

- `smiles`: molecule SMILES
- `doi`: reference DOI
- `inchlKEY`: InChIKey (**kept as-is from the original file**, including spelling)
- `construction`: material category (e.g., polymer)
- `architecture`: device architecture (e.g., bulk)
- `complement`: complementary material (e.g., PC61BM/PC71BM)
- `HOMO`, `LUMO`: energy levels (commonly in eV; values typically negative)
- `electrochemical_gap`, `optical_gap`: gap-related quantities (commonly in eV)
- `PCE`: power conversion efficiency (commonly %)
- `V_OC`: open-circuit voltage (commonly V)
- `J_SC`: short-circuit current density (commonly mA/cm²; see original file for units)
- `fill_factor`: fill factor (raw data may store it in percent-like values)

Missing values can appear as `nan`.

### `PFD.csv`
Header:
`Nickname,PCE_max(%),PCE_ave(%),Voc,Jsc,FF,Mw,Mn,PDI,Monomer,HOMO,LUMO,bandgap,SMILES`

- `Nickname`: material/polymer nickname
- `PCE_max(%)`, `PCE_ave(%)`: max/average PCE (percent)
- `Voc`, `Jsc`, `FF`: device metrics
- `Mw`, `Mn`, `PDI`: molecular weight related fields
- `Monomer`: monomer-related field (kept as-is)
- `HOMO`, `LUMO`, `bandgap`: electronic properties
- `SMILES`: structure SMILES

### `NFA.csv` and `PD.csv`
Both share the same header (kept exactly as in the original CSV, including quotes/hyphens):
`PCE_max(%),PCE_ave(%),Jsc(mA/cm2),FF,Voc(V),HOMO_n(eV),'-LUMO_n(eV),Eg_n(eV),n(SMILES),M (g/mol),HOMO_n(eV),'-LUMO_n(eV),Eg_n(eV),p(SMILES),Mw (kg/mol),Mn(kg/mol),PDI`

This can be interpreted as a paired-material record (n / p) with device performance:
- `PCE_max(%)`, `PCE_ave(%)`, `Jsc(mA/cm2)`, `FF`, `Voc(V)`: device performance
- First group (`... n(SMILES)`): n-side material properties and SMILES
- `M (g/mol)`: molecular mass field (unit in header)
- Second group (`... p(SMILES)`): p-side material properties and SMILES  
  (column names are not renamed in the raw file)
- `Mw (kg/mol)`, `Mn(kg/mol)`, `PDI`: molecular-weight related fields (units in headers)


## How to Use
This repository provides raw CSVs. Example:

```python
import pandas as pd

df_cepdb = pd.read_csv("CEPDB.csv")
df_hopv = pd.read_csv("HOPV.csv")
df_pfd  = pd.read_csv("PFD.csv")
df_nfa  = pd.read_csv("NFA.csv")
df_pd   = pd.read_csv("PD.csv")
```

For **ring-graph construction** and **training**, use the scripts in the RingFormer GitHub repository (e.g., `generate_ring_graphs.py`, `train.py` as described in the upstream README).

## Citation
If you use this dataset, please cite:

```bibtex
@inproceedings{ding2025ringformer,
  title={RingFormer: a ring-enhanced graph transformer for organic solar cell property prediction},
  author={Ding, Zhihao and Zhang, Ting and Li, Yiran and Shi, Jieming and Zhang, Chen Jason},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={1},
  pages={155--163},
  year={2025}
}
```

## License
This repository republishes the original CSV tables used by RingFormer for reproducibility and re-processing. Please refer to the upstream RingFormer GitHub repository for licensing/usage terms of the data. If the upstream repository does not clearly specify a license for the data, please verify compliance before downstream use.