Tools
Property Predictor
Four machine-learning models, trained on experimental data, predicting properties from structure alone. Every number here is a prediction, not a measurement.
Predicted
A prediction from structure, not a measured value. It carries no uncertainty estimate and no check that your structure resembles anything the model was trained on. Treat it as a starting point for an experiment, never as data.
About these models
The four models were developed at Polymat, the Basque Center for Macromolecular Design and Engineering, and are described in the two papers below. The reactivity-ratio network was trained on more than 5,000 monomer pairs and outperforms the classical Q–e scheme; the other three came out of a later framework aimed at assessing bio-based monomers as replacements for conventional ones.
No predicted value appears anywhere else in the Atlas. Measured, cited properties and model output are kept apart on purpose.
Papers
- [1]An artificial neural network to predict reactivity ratios in radical copolymerizationPolymer Chemistry 2023, 14, 2779--2787https://doi.org/10.1039/D3PY00246B[farajzadehahary-2023-reactivity-ratios]
- [2]Multi-Property Machine Learning Models to Accelerate the Transition Toward Bio-Based Emulsion PolymersAdvanced Intelligent Discoveryhttps://doi.org/10.1002/aidi.202500248[farajzadehahary-2026-multiproperty-models]