skill Proprietary (Paramus) — guidance only; execution performed by Paramus tools under their own license gating.
Screen and rank candidate polymers by predicted glass transition temperature (Tg) against a target value or window, orchestrating BigSMILES validation, polymer fingerprinting, and the Paramus BRAIN Tg model.When to use: a user provides repeat units (BigSMILES/SMILES) and asks which best hit a Tg target.Multi-tool WORKFLOW, not one call.

Paramus — Polymer Tg Screening

Overview

Validate structures, featurize them, predict Tg with the deterministic BRAIN model, and rank candidates by fit to a target. The value is the end-to-end screen with a materials-design decision rule — not any single prediction call.

Guidance vs execution: predictions come from the tested BRAIN Tg model with version + uncertainty attached. Do not fit or estimate Tg with hand-written code — route to the model.

When to Use

  • User supplies polymers (BigSMILES/SMILES) and asks which match a target Tg
  • Ranking monomer/copolymer options for a thermal target Do not use for: measured-data curve fitting, or properties other than Tg.

Workflow

Candidates + target Tg (or window)
  ↓ 1. Validate & normalize BigSMILES   → Paramus polymer parsing tool
  ↓ 2. Polymer fingerprint / features   → Paramus fingerprinting tool
  ↓ 3. Predict Tg (with uncertainty)    → Paramus BRAIN Tg model
  ↓ 4. Score vs target                  → distance / inside-window test
  ↓ 5. Rank & report                    → ranked table + model version + uncertainty

Procedure

  1. Discover tools via search ("BigSMILES validation", "polymer fingerprint", "Tg prediction"); confirm with get_schema.
  2. Validate every BigSMILES; report malformed inputs explicitly — never silently skip.
  3. Featurize the validated set (batched if supported) so all share one feature convention.
  4. Predict Tg; capture predicted Tg + uncertainty + units + model version.
  5. Score against target (window → inside/outside; point → |Tg_pred − Tg_target|). State the rule.
  6. Report: polymer, Tg ± uncertainty, fit-to-target, rank, model version.

Domain Judgment

  • Uncertainty gates ranking: an error bar straddling the window boundary is "uncertain-fit", not #1.
  • Extrapolation warning: structures far from the model's applicability domain → demote confidence.
  • Never mix predictions across model versions in one ranking.

Fallbacks

  • BRAIN Tg model unavailable → report; offer nearest thermal-property tool. No local group-contribution hack.
  • Endpoint unreachable → stop and report; do not fall back to local prediction code.