Oxidation State Prediction
What This Tool Does
Predict likely element oxidation states for formulas or CIF structures using BERTOS, pymatgen, oxi.matr.io PNAS, or TOSS.
Typical Inputs
- Formula list in the text box.
- Optional CSV/TXT file using the first column or line-delimited formulas.
- Optional CIF files, ZIP archives, TAR archives, TAR.GZ, or TGZ archives.
- Oxidation-state method: BERTOS, pymatgen oxidation-state guess, oxi.matr.io PNAS electrochemical series, TOSS structure-based prediction, or Compare all methods.
Methods
- BERTOS oxidation-state prediction: runs the configured BERTOS transformer checkpoint when available.
- pymatgen oxidation-state guess: uses
Composition.oxi_state_guesses()for common chemistry-aware assignments. - oxi.matr.io PNAS electrochemical series: uses a local Python port of the TRI-AMDD oxidation-state-api-public composition/likelihood model.
- TOSS structure-based prediction: runs the local TOSS GNN CIF-based oxidation-state predictor on uploaded structures, with the traditional MAP route kept as a fallback when its generated artifacts are available.
- Compare all methods: runs BERTOS, pymatgen, PNAS, and TOSS side-by-side. For formula-only rows, TOSS is marked not applicable; for CIF rows, formulas are extracted with pymatgen for the composition methods and the raw CIF is sent to TOSS.
PNAS citation: Mueller, Tim, Joseph Montoya, Weike Ye, Xiangyun Lei, Linda Hung, Jens Hummelshøj, Michael Puzon, Daniel Martinez, Chris Fajardo, and Rachel Abela. "An electrochemical series for materials." Proceedings of the National Academy of Sciences 121, no. 38 (2024): e2320134121.
TOSS citation: Yin, Yue, and Hai Xiao. "Oxidation states in solids from data-driven paradigms." Chemical Science 16, no. 42 (2025): 19917-19928.
Output
- Predicted oxidation states by element.
- Net charge and charge-neutral flag when available.
- PNAS likelihood and mapped potential for oxi.matr.io runs.
- TOSS site-level oxidation states and coordination numbers when a CIF structure run succeeds.
- Downloadable CSV result table.
- In Compare all methods mode, one row per formula/CIF with BERTOS, pymatgen, PNAS, and TOSS columns side by side.
Limitations
Formal oxidation states are model assignments. They can be wrong for unusual chemistries, metallic/covalent bonding, mixed valence, disorder, nonstoichiometry, and formulas outside a model's domain. TOSS requires the local TOSS code, Python dependencies, and pretrained GNN model artifacts to be available on the backend. If that installation is incomplete, the result table marks TOSS as unavailable instead of failing the whole batch. Use method disagreement as a cue for closer inspection.
Acknowledgements
This tool uses and acknowledges the following oxidation-state prediction methods and references:
- BERTOS: Fu, Nihang, Jeffrey Hu, Ying Feng, Gregory Morrison, Hans-Conrad zur Loye, and Jianjun Hu. "Composition based oxidation state prediction of materials using deep learning language models." Advanced Science 10, no. 28 (2023): 2301011.
- OXI: Mueller, Tim, Joseph Montoya, Weike Ye, Xiangyun Lei, Linda Hung, Jens Hummelshøj, Michael Puzon, Daniel Martinez, Chris Fajardo, and Rachel Abela. "An electrochemical series for materials." Proceedings of the National Academy of Sciences 121, no. 38 (2024): e2320134121.
- TOSS: Yin, Yue, and Hai Xiao. "Oxidation states in solids from data-driven paradigms." Chemical Science 16, no. 42 (2025): 19917-19928.