A new tool for designing polymer electrolytes

PolyGPT generates polymer repeat units based on specific property targets and chemical features.

Researchers developed PolyGPT, a Transformer with 38.3 million parameters. This tool generates polymer repeat units. It uses optional property targets and electrolyte-domain conditions.

The model can include or exclude specific chemical features. A polymer repeat unit is the basic building block of a polymer. A glass-transition temperature, or TgT_g, is a property target. Electrolyte-domain conditioning refers to requests for electrolyte-associated structures.

Unconditional generation achieved a structural validity of 99.46%. Electrolyte-only generation achieved a structural validity of 99.59%. However, electrolyte conditioning increased repetition. Uniqueness among valid outputs fell from 99.99% to 51.74%.

At a TgT_g target of 246.11 K, adding the electrolyte condition reduced the predictor-estimated mean absolute error from 26.97 to 26.07 K. But the fraction of attempts yielding distinct valid polymers within 50 K of the target fell from 82.35% to 32.31%.

Domain conditioning also weakened control of inclusion requests. The mean fraction of attempts producing valid polymers with the requested chemical feature fell from 94.27% to 31.27%. Property error alone is insufficient for evaluating conditional polymer generation.

Evaluation must consider repetition and training-catalog novelty. It must also consider if combined design requests are satisfied.

Key facts

  • PolyGPT has 38.3 million parameters.
  • Electrolyte conditioning increased repetition and reduced uniqueness from 99.99% to 51.74%.
  • Adding electrolyte conditions reduced the mean absolute error for a TgT_g target of 246.11 K from 26.97 to 26.07 K.

All sections

Edition