Molecular Structure
Canonical structure
No structure evaluated
MW --cLogP --TPSA --
Design Objective Comparison
Optical thresholds, not an applicability domain
No prediction available.
Emission deficit: --Stokes shift deficit: --
Ensemble-Member Dispersion
Standard deviation across seed models
Emission SD: --Stokes shift SD: --
Member dispersion is not a calibrated confidence interval or a probability of prediction correctness. Molecular weight and cLogP are descriptive properties, not generic reliability penalties.
Optimization Protocol
Complete-probe optical prediction
- Emission objective
- ≥ 700 nm
- Stokes shift objective
- ≥ 100 nm
- Editable domain
- Fluorophore fragment only
- Fixed domain
- All linker and recognition-context atoms
- Attachment protection
- Radius 1 around the fluorophore attachment
- Target stopping quota
- 10 distinct new target probes
- Review shortlist
- Up to 10; earlier rounds first
- Termination
- Round budget, target quota, or exhausted branches
The two mapped fragments must reconstruct the complete probe exactly. The fixed context is not, by itself, evidence of analyte-specific recognition.
Optical estimates, module preservation, and configured feasibility limits are separate criteria. Target attainment does not establish experimental performance.
Earlier target-generation rounds take priority. Optical Pareto ranking is applied within each round; diversity resolves selection within the cutoff round. All target hits remain archived, including those beyond the stopping quota.
No optimization run submitted.
Candidate Probes
No candidates available.
| Candidate | Review / archive rank | Generated round | First target round | Search stop round | Emission / nm | Stokes / nm | Optical score | Cumulative FP distance | Cumulative absolute HA change | Target | Pathway |
|---|---|---|---|---|---|---|---|---|---|---|---|
| No optimization results available. | |||||||||||
Molecular Editing Pathway
Recorded parent-child edits
- No candidate selected.
Round Audit
Generation, constraint checks and retained structures
| Round | Generated | Structure pass | Feasibility pass | Retained | New targets | Termination |
|---|---|---|---|---|---|---|
| No round records available. | ||||||
Probe Definition Contract
Same scientific protocol as individual probe optimization
- Complete probe
full_probe_smiles- Editable fluorophore
fluorophore_fragment
One attachment atom [*:1]- Fixed context
recognition_fragment
One attachment atom [*:2]; linker and recognition atoms preserved- Maximum rounds
max_rounds
Integer from 1 to 5, per probe- Solvent
solvent_smiles
Water when omitted or blank- Probe identifier
probe_id
Unique within the batch; optional- Source record
reference_probe_id
Optional verified record with its retained feasibility limits
Mapped fragments reconstruct the complete probe exactly. Optical targets, editing rules and feasibility criteria remain unchanged. Completion does not imply target attainment.
Per-Probe Optimization Record
Baseline properties and the candidate selected by the retained optimization protocol
| Row / probe | Status | Maximum / completed rounds | Initial emission / Stokes nm | Selected emission / Stokes nm | Retained / target probes | Selected candidate | generated / first target / stop round | Record |
|---|---|---|---|---|---|---|---|
| No batch optimization records available. | |||||||
Probe Record
Candidate Probes
| Candidate | Review / archive rank | Generated round | First target round | Search stop round | Emission / nm | Stokes / nm | Optical score | Cumulative FP distance | Cumulative absolute HA change | Target | Pathway |
|---|
Molecular Editing Pathway
Round Audit
Generation, feasibility checks and retained structures for the selected probe
| Round | Generated | Structure pass | Feasibility pass | Retained | New targets | Termination |
|---|
Model Configuration
Local artifact package
Scientific Scope
Property prediction in support of molecular design
The model estimates emission maximum and Stokes shift for molecular-solvent pairs. NIR-I emission and large Stokes shift are optimization objectives, not restrictions on general property prediction.
Historical frozen-test performance uses the original saved feature matrix and pair split. It is not a new evaluation of online inference, a molecule-held-out validation, or experimental evidence.
The prediction model, preprocessing, edit rules, SHAP motifs and feasibility criteria are retained from the underlying scientific workflow without retraining or additional algorithms.
Historical Frozen-Test Performance
Archived model evaluation
| Model / seed | Emission MAE / nm | Emission R² | Stokes MAE / nm | Stokes R² |
|---|