Paramus — Synthetic Accessibility Screen

Overview

Rank candidates by how realistic they are to make: SA score, problematic-motif flags, and complexity estimates fused into a make-ability tier. The value is the feasibility judgment, not a lone SA score.

Guidance vs execution: SA scores and motif detection come from tested Paramus tools. Do not estimate synthesizability from model intuition alone.

When to Use

• Filtering a virtual/generated set to synthesizable candidates before assay/purchase
• Prioritizing which designs to route to chemistry Do not use for: full retrosynthesis route planning (hand off to the retrosynthesis workflow/tool).

Workflow

Candidate set (SMILES)
  ↓ 1. Parse & standardize        → Paramus cheminformatics tool
  ↓ 2. SA score                   → Paramus synthetic-accessibility tool
  ↓ 3. Motif / complexity flags   → strained rings, unusual valences, stereocenters count
  ↓ 4. Tier by make-ability       → Easy / Feasible / Hard
  ↓ 5. Report                     → ranked table + flags + provenance

Procedure

1. Discover tools via search/get_schema ("synthetic accessibility", "structural alerts").
2. Standardize inputs; quarantine parse failures.
3. Compute SA scores; capture method + version.
4. Flag problematic motifs and count stereocenters/ring complexity.
5. Tier with an explicit rule; report ranked candidates with flags and rationale.

Domain Judgment

• SA score is a heuristic, not a route — a "hard" flag warrants a chemist's look, not an auto-drop.
• Many stereocenters/strained systems raise cost even at a decent SA score.
• Separate "cannot buy building blocks" from "hard to assemble" where the tool distinguishes them.

Fallbacks

• SA tool unavailable → deliver motif/complexity flags only, clearly scoped.
• Endpoint unreachable → stop and report; no local SA heuristics.