---
license: mit
library_name: chemprop
tags: [chemistry, thermodynamics, melting-point, enthalpy-of-fusion, solubility, chemprop, screening-properties]
---

# Fusion-Cycle: melting temperature and enthalpy of fusion

These are two five-model [Chemprop](https://github.com/chemprop/chemprop) 2 ensembles from **Fusion-Cycle**:

- `trained_models/MP`: melting temperature, in K
- `trained_models/dHfus`: enthalpy of fusion, in **kcal/mol**

Both take a SMILES string as their only input. The files are byte-identical copies from the upstream repository. They are re-hosted here so that screening-properties (the property package of the ScreeningTool process-screening framework) can download them on first use. The solubility (γ) and mixture models of Fusion-Cycle are not included.

**All credit goes to the original authors.**

- Code and weights: [emadalibrahim/Fusion-Cycle](https://github.com/emadalibrahim/Fusion-Cycle), commit `55d8ca577ac1cb5aeea2685523223bd5c9c19977`.
- Licence: the upstream package metadata (its conda recipe) declares **MIT**. The upstream repository contains no separate LICENSE file. If you are an author and want this re-hosting changed, please open a discussion on this repository.

> Al Ibrahim, Emad, et al. "Accurately predicting solubility curves via a thermodynamic cycle, machine learning, and solvent ensembles." *Journal of the American Chemical Society* 147.49 (2025): 45057-45069.

## Use

```python
from chemprop import models
mpnn = models.MPNN.load_from_file("trained_models/MP/model_0.pt")  # pickle: only load files you have verified
```

screening-properties averages the five members, converts ΔHfus to J/mol (× 4184), and checks every file's SHA-256 before loading. It does not predict molecules without carbon or with more than one fragment (salts), because those lie outside the organic training domain.