Publication WebServer Synthesis-aware molecular assessment

From molecular idea
to a synthesis-aware
score.

GESA estimates the probability that a molecular structure belongs to the hard-to-synthesize class by fusing reaction-informed, task-specific and explicit structural evidence.

CLASSIC SYNTHESIS / ACETYLATIONASSEMBLING ASPIRIN

01StandardizeCanonical, stereochemistry-aware SMILES

02PredictStable GESA score, P(HS)

03PlanOptional Retro* route search

01Submit molecules

GESA INPUT

Paste a valid SMILES. It does not need to be canonical—the server will normalize and verify it before inference.

Try an example

0.0Lower difficulty signal

1.0Higher difficulty signal

✓GESA score is P(HS), not reaction yield or proof of laboratory feasibility. The released classification threshold is 0.50.

GESA / BATCHBATCH PREDICTION

Preparing CSV0.0 s elapsed

GESA / RESULTCANONICAL MOLECULE

01 / STANDARDIZED STRUCTURE

INPUT

CANONICAL

02 / SYNTHETIC ACCESSIBILITY

0.000P(HS)

Easier to synthesize

The model signal falls below the released HS decision threshold.

LOWER SIGNALTHRESHOLD 0.50HIGHER SIGNAL

DECISION MARGIN —

Distance from the 0.50 classification boundary.

MW—g/mol

LogP—octanol/water

TPSA—Å²

RotB—rotatable bonds

Rings—ring count

Heavy—heavy atoms

WHAT IT MEANS

A model-based difficulty signal

The score ranks the submitted structure on the GESA hard-to-synthesize axis. Compare it with 0.50 for the released ES/HS classification.

WHAT IT DOES NOT MEAN

Not a yield or guarantee

Reagents, conditions, selectivity, availability and experimental execution are not fully represented by a single accessibility score.

RECOMMENDED NEXT STEP

Inspect a candidate route

Use the optional Retro* search below to test whether the model can connect this target to its configured starting-material library.

03 / OPTIONAL RETROSYNTHESIS

Move from score to a candidate route.

Retro* runs a neural-guided A* search using the same canonical target. A zero-step result means a match in the configured starting-material library, not confirmed current commercial stock.

ABOUT THE PROJECT

About GESA

GESA (Generative-space Expanded Synthetic Accessibility) is a research framework and publication companion WebServer developed by the Laboratory of Molecular Modeling and Drug Design at East China University of Science and Technology.

01 / PLATFORM OVERVIEW

Synthesis-aware molecular assessment

GESA estimates the probability that a molecular structure belongs to the hard-to-synthesize class. It is designed to complement molecular generation and prioritization with an interpretable accessibility signal.

The WebServer standardizes molecular input, performs deterministic inference and optionally connects the target to a candidate Retro* route.

02 / MODEL ARCHITECTURE

Multi-view evidence fusion

Task-GAT+Reaction-GINE→Atom-wise gating→GESA score

Task-specific and reaction-informed graph encoders are fused before attentive molecular readout and combined with fourteen explicit synthetic-accessibility priors.

03 / EXTERNAL EVALUATION

Validated across four test sets

0.983TS1 AUROC

0.941TS2 AUROC

0.884TS3 AUROC

0.901TS4 AUROC

The released public checkpoint was selected using internal validation rather than the external TS1–TS4 test sets.

AFFILIATIONSResearch environment

LMMD GroupLaboratory of Molecular Modeling and Drug Design ↗ School of PharmacyEast China University of Science and Technology ↗ ECUSTEast China University of Science and Technology ↗

METHOD

Task-GAT + Reaction-GINE → atom-wise source gating → attentive molecular readout + 14 explicit SA priors.

GESA · Generative-space Expanded Synthetic Accessibility