Hybrid Machine Learning and Graph Neural Network Approaches for Accurate Melting Point Prediction of Organic Crystals
Received: 24 Oct 2026 Revised: 09 Nov 2026 Accepted: 09 Nov 2026
Published: 2026, vol. 30, issue 4, pp. 490–493
Abstract
Melting point prediction is crucial for chemistry and pharmaceuticals, but current methods face accuracy and efficiency challenges. We curated a harmonized dataset of 117,752 organic crystals. We compared machine learning models using molecular embeddings and graph neural networks. A hybrid TabM model with concatenated ChemBERTa and MoLFormer embeddings achieves MAE of 26.75 K in 27 minutes of training, offering an optimal balance of accuracy and computational cost compared to Chemprop (MAE 26.44 K, 45 minutes).
Keywords: melting point, organic crystals, machine learning, graph neural networks, ChemBERTa, MoLFormer, TabM, Chemprop.
BibTeX
@article{IS-Dubonos-Novikov-Bokov-Mitrofanov2026,
author = {Dubonos, Nikolai Sergeevich and Novikov, Alexander Sergeevich and Bokov, Grigoriy Vladimirovich and Mitrofanov, Artem Alexandrovich},
title = {{Hybrid Machine Learning and Graph Neural Network Approaches for Accurate Melting Point Prediction of Organic Crystals}},
journal = {Intelligent Systems. Theory and Applications},
year = {2026},
volume = {30},
number = {4},
pages = {490--493},
}
AMSBIB
\Bibitem{IS-Dubonos-Novikov-Bokov-Mitrofanov2026}
\by N.\,S.~Dubonos, A.\,S.~Novikov, G.\,V.~Bokov, A.\,A.~Mitrofanov
\paper Hybrid Machine Learning and Graph Neural Network Approaches for Accurate Melting Point Prediction of Organic Crystals
\jour Intelligent Systems. Theory and Applications
\yr 2026
\vol 30
\issue 4
\pages 490--493
\lang In Russian
RU