Synthesizability score (SynScore)

What it does: give a target molecule a continuous synthesizability score (SynScore) for ranking a set of molecules by how easy they are to make. It runs one route planning search internally (the purchasable library is switchable and customizable — see building-block libraries), then folds the outcome into a score. Definition and operating point: Research · Synthesizability Score; install: Install & Overview.

\[\mathrm{SynScore} = \frac{1}{(U+1)^{U}}\]

U = the number of non-purchasable starting materials in the best route: all purchasable (U=0) → 1.0, U=1 → 0.5, U=2 → 0.11, U=3 → 0.016, no route at all → 0. It falls off sharply, so it cleanly separates "fully solved / a few blocks missing / many missing".

Command line

# defaults to the simplification-constrained model @ expansion width k=10 (recommended);
# --original switches to the unconstrained model
synomega score --targets targets.smi --out scores.jsonl

Python

import synomega

scorer = synomega.load_default_scorer()              # default simplify=True, k=10
r = scorer.score("CC(=O)Nc1ccccc1O", max_steps=5)
print(r.as_dict())
# {'smiles': ..., 'solved': True, 'score': 1.0, 'bb_coverage': 1.0,
#  'min_steps': 2, 'min_route_depth': 2, 'num_leaves': 2,
#  'num_purchasable_leaves': 2, 'expansions': ..., 'terminated_by': 'solved', ...}

# batch: one SMILES per line
report = scorer.score_batch(open("targets.smi").read().split())
print(report.solve_rate, report.mean_bb_coverage)
print(report.describe())
df = report.to_dataframe()                            # one row per molecule (needs pandas)

A single target returns a MoleculeReport (.score / .solved / .bb_coverage / .min_steps / .num_unpurchasable_leaves (= U) / .as_dict()); a batch returns a BatchReport (.solve_rate / .mean_bb_coverage / .describe() / .to_dataframe() / .to_json()).

Reading the score

Case	U	SynScore
all starting materials purchasable (solved)	0	1.0
one block short	1	0.5
two short	2	0.11
no route at all	—	0

solved is the binary "is there an all-purchasable route within the depth", for comparing to published solve_rate; SynScore is the continuous, near-miss-aware, rankable version for ordering a set of molecules.

Parameters

Parameter (CLI / Python)	Default	Meaning
--targets	required	one SMILES per line
--original / simplify=	simplify model	score with the unconstrained single-step model instead
--max-steps / max_steps=	5	route depth cap
--exclude-target	off	treat the target as not purchasable (avoid a buyable target scoring 1.0 in zero steps)