EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

ReactionFormer: A Controlled Benchmark for Infrared Spectral Reaction Monitoring

Sergio Josè Peresson, Danilo Croce, Roberto Basili

ACS Physical Chemistry Au · 20.11.2026

DOI: 10.1021/acsphyschemau.6c00130 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ACS Physical Chemistry Au, online erschienen am 20.11.2026. Verfasst von Sergio Josè Peresson und 2 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Reaction monitoring is central to chemical process development, but systematic evaluation of machine-learning models is limited by the scarcity and heterogeneity of experimental monitoring data. ReactionFormer provides a controlled and reproducible framework for evaluating machine-learning-based reaction-state inference from simulated infrared mixture spectra and molecular reaction context. The benchmark pairs reaction context with simulated spectral snapshots generated from computed and experiment-derived pure-component infrared spectra. The reaction-monitoring trajectories are not experimentally measured; they are simulated from pure-component spectra and represented as serialized sequences combining reactant and product simplified molecular-input line-entry system (SMILES) strings with discretized infrared mixture spectra. ReactionFormer supports reaction-progress regression, practical-completion classification, spectrum-only ablation, cross-source transfer, and low-data adaptation under matched reaction-level splits. The results show that infrared snapshots contain substantial reaction-state information, molecular context provides additional gains, and masked-language-model pretraining improves cross-source transfer and adaptation when only a limited number of labeled examples from the experiment-derived spectral variant are available. Beyond predictive performance, the study quantifies how molecular context, spectral-source variation, and self-supervised pretraining affect the robustness and transferability of learned spectroscopic representations.

Zitieren

APA

Peresson, S. J., Croce, D., & Basili, R. (2026). ReactionFormer: A Controlled Benchmark for Infrared Spectral Reaction Monitoring. ACS Physical Chemistry Au. https://doi.org/10.1021/acsphyschemau.6c00130

Vancouver

Peresson SJ, Croce D, Basili R. ReactionFormer: A Controlled Benchmark for Infrared Spectral Reaction Monitoring. ACS Physical Chemistry Au. 2026. doi:10.1021/acsphyschemau.6c00130

BibTeX
@article{Peresson2026,
  author = {Peresson, Sergio Josè and Croce, Danilo and Basili, Roberto},
  title = {ReactionFormer: A Controlled Benchmark for Infrared Spectral Reaction Monitoring},
  journal = {ACS Physical Chemistry Au},
  year = {2026},
  doi = {10.1021/acsphyschemau.6c00130}
}
Kostenlos starten →