We use three multitask deep neural networks to predict the thermal,
mechanical, and gas permeability properties of polymers.
Polymer structures are represented as SMILES strings
that use two stars to indicate the two endpoints of the
repetitive unit of the polymers, but otherwise follow the SMILES syntax.
See PolymerGenome Guide and
OpenSmiles for more information.
The shaded bands represent $$1\sigma$$ uncertainty estimates using
Monte Carlo dropouts.