rewire.itbenchmarks

← Back to use cases

Proteins and complexes · Use case

Assess rhodopsin wavelength prediction across sequence backgrounds

Which sequence-based configurations merit testing before selecting rhodopsins with different absorption wavelengths in a new wild-type background?

Your decision and inputs

Use the existing held-out-background evaluations to choose controls and exact model configurations for a new, prospectively held-out spectral-tuning experiment.

Who this is for
Protein engineers planning rhodopsin spectral-tuning experiments; Computational researchers checking transfer to new sequence backgrounds
Context
Research
Inputs
  • Complete rhodopsin amino-acid sequences
  • Training measurements of peak absorption wavelength and frozen assignments that separate wild-type backgrounds
Expected output
Evidence for the exact frozen sequence probes and controls, plus the validation still needed for a new background; no calibrated wavelength or wet-lab success guarantee.
Biological setting
One complete FLIP2 Rhomax by_wild_type test split: 584 training, 116 unused validation and 184 test records. The experimental endpoint is peak absorption wavelength in nanometres.

Outside this use case

  • Rhodopsin activation efficiency, expression, photostability and cellular function
  • Zero-shot likelihood scoring, alternative pooling or fine-tuning not evaluated in these runs
  • General protein fitness, all FLIP2 landscapes and clinical use

What this establishes for clinical research

Research only. Absorption-wavelength ranking does not establish suitability for an optogenetic intervention, patient care or treatment.

Which evaluations inform this question?

Evidence is grouped by its protocol. Relevance refers to the stated endpoint and context; it is separate from clinical validation and from the review method.

FLIP2 Rhomax by_wild_type

Current source-reviewed mapping

Proxy evidence: transfer to this question is limited

The archived background split directly tests one transfer setting; choosing constructs in a new experiment requires validation beyond those 184 rows.

Assessed endpoint
Spearman rank association and full-ranking NDCG for absorption-wavelength predictions on 184 held-out Rhomax sequences.
Evaluation protocol
FLIP2 Rhomax by_wild_type
Computational task
A reviewed task relationship is not recorded for this protocol.
Input and population constraints
  • The same frozen Rhomax by_wild_type assignments cover 584 training, 116 validation and 184 test records. Validation labels do not select these fixed settings.
  • Two exact ESM-2 checkpoints use frozen final-layer residue means, no MSA or templates, and a train-only alpha-10 ridge probe with train-only target scaling. They are not zero-shot likelihood scorers.
  • The 40-feature and 22-feature composition controls are distinct configurations. Training mean is the existing constant-control evaluation; repeated control checks do not add replications.

Limits on interpretation

  • A spectral-tuning endpoint is not opsin activation efficiency, expression, photostability, cellular function, general protein fitness or clinical usefulness.
  • Spearman and full-ranking NDCG do not establish wavelength calibration, top-k precision or a prospective experimental hit rate. Constant predictions have undefined Spearman; their NDCG is a control value, not strong predictive evidence.
  • No intervals or seed-variability estimates. The 35M checkpoint was chosen after seeing the 8M outcome; the source describes an exploratory follow-up, not a preregistered family-scale comparison.
  • ESM-2 pretraining overlap is unresolved. A frozen linear probe result does not establish the value of alternative pooling, fine-tuning or every model-family configuration.
  • Recorded timing covers sections of execution and is not a hardware-normalised deployment-cost comparison. Human domain review and external replication are not established.

Automated source review · 2026-09-28 · Codex research curation

Primary-source curation and separate automated cross-review checked exact evidence identities, comparator coverage, endpoint relevance and transfer limits. No new execution, human scientific review, independent replication or clinical validation.

Evaluated configurations

Each configuration below belongs to this protocol. Inspect its inputs, population and scoring conditions before comparing it with another evaluation.

Amino-acid composition + fixed ridge (FLIP2 Rhomax by_wild_type)

Rewire evaluation · Source checked

Predict measured Rhomax wavelengths (nm) from amino-acid sequence on all 184 held-out archived test records. This selected split is not a complete FLIP2 score.

Population and split
184/184 · canonical test
Inputs and adaptation
Complete source sequence only; no assay labels during prediction · train-only fitting
Evaluation budget
one local CPU evaluation; no hyperparameter search outside training
Runtime and memory

Inference and fit section: 0.012 s.

Metric evaluation section: 0.0109 s.

Device: Not reported. Batch size: 32.

These timings describe the recorded sections of this run, not total runtime or a general hardware benchmark. Peak memory is not reported.

Recorded results for this configuration
MetricValueCoverageUncertainty and source
ndcg0.955

dimensionless · higher

184/184

Not reported

Result provenance
spearman0.418

dimensionless · higher

184/184

Not reported

Result provenance

Uncertainty: Not estimated; one complete selected evaluation.

Evaluation methods, evidence and reproduction

Recipe: generate and evaluate predictions

This committed script regenerates the selected local evaluation with the recorded inputs and configuration. Sequence-control instructions execute all four controls; select this evaluation from their outputs. Raw prior predictions are private, so public evidence alone cannot rescore them.

Training-mean control (FLIP2 Rhomax by_wild_type)

Rewire evaluation · Source checked

Predict measured Rhomax wavelengths (nm) from amino-acid sequence on all 184 held-out archived test records. This selected split is not a complete FLIP2 score.

Population and split
184/184 · canonical test
Inputs and adaptation
Complete source sequence only; no assay labels during prediction · train-only fitting
Evaluation budget
one local CPU evaluation; no hyperparameter search outside training
Runtime and memory

Inference and fit section: 0.000942 s.

Metric evaluation section: 0.0165 s.

Device: Not reported. Batch size: 32.

These timings describe the recorded sections of this run, not total runtime or a general hardware benchmark. Peak memory is not reported.

Recorded results for this configuration
MetricValueCoverageUncertainty and source
ndcg0.921

dimensionless · higher

184/184

Not reported

Result provenance
spearmanundefined

dimensionless · higher

184/184

Not reported

Result provenance

Uncertainty: Not estimated; one complete selected evaluation.

Evaluation methods, evidence and reproduction

Recipe: generate and evaluate predictions

This committed script regenerates the selected local evaluation with the recorded inputs and configuration. Sequence-control instructions execute all four controls; select this evaluation from their outputs. Raw prior predictions are private, so public evidence alone cannot rescore them.

Protein composition + fixed ridge probe (FLIP2 Rhomax by_wild_type)

Rewire evaluation · Source checked

Frozen features with a fixed alpha-10 ridge head fitted only on training rows; all 184 Rhomax by_wild_type held-out proteins scored. This selected split is not a whole FLIP2 score.

Population and split
184/184 · canonical test
Inputs and adaptation
protocol allowlisted biological inputs; see baseline configuration for extra inputs · fixed train-only ridge on frozen representations
Evaluation budget
one local CPU evaluation; no held-out hyperparameter selection
Runtime and memory

Inference and fit section: 0.0105 s.

Metric evaluation section: 0.0133 s.

Device: Not reported. Batch size: 32.

These timings describe the recorded sections of this run, not total runtime or a general hardware benchmark. Peak memory is not reported.

Recorded results for this configuration
MetricValueCoverageUncertainty and source
ndcg0.955

dimensionless · higher

184/184

Not reported

Result provenance
spearman0.418

dimensionless · higher

184/184

Not reported

Result provenance

Uncertainty: Not estimated; one selected evaluation.

Evaluation methods, evidence and reproduction

No execution recipe has been verified for this exact configuration and evaluation. Inspect its methods and original run documentation before attempting reproduction.

ESM-2 35M frozen residue-mean embeddings + fixed ridge (Rewire) (FLIP2 Rhomax by_wild_type)

Rewire evaluation · Source checked

Frozen features with a fixed alpha-10 ridge head fitted only on training rows; all 184 Rhomax by_wild_type held-out proteins scored. This selected split is not a whole FLIP2 score.

Population and split
184/184 · canonical test
Inputs and adaptation
Complete amino-acid sequences; no labels enter encoder; no MSA/templates · fixed train-only ridge on frozen representations
Evaluation budget
one local CPU evaluation; no held-out hyperparameter selection
Runtime and memory

Inference and fit section: 64.1 s.

Metric evaluation section: 0.162 s.

Device: cpu. Batch size: 4.

These timings describe the recorded sections of this run, not total runtime or a general hardware benchmark. Peak memory is not reported.

Recorded results for this configuration
MetricValueCoverageUncertainty and source
ndcg0.907

dimensionless · higher

184/184

Not reported

Result provenance
spearman-0.222

dimensionless · higher

184/184

Not reported

Result provenance

Uncertainty: Not estimated; one selected evaluation.

Evaluation methods, evidence and reproduction

No execution recipe has been verified for this exact configuration and evaluation. Inspect its methods and original run documentation before attempting reproduction.

ESM-2 8M frozen residue-mean embeddings + fixed ridge (Rewire) (FLIP2 Rhomax by_wild_type)

Rewire evaluation · Source checked

Frozen features with a fixed alpha-10 ridge head fitted only on training rows; all 184 Rhomax by_wild_type held-out proteins scored. This selected split is not a whole FLIP2 score.

Population and split
184/184 · canonical test
Inputs and adaptation
Complete amino-acid sequences; no labels enter encoder; no MSA/templates · fixed train-only ridge on frozen representations
Evaluation budget
one local CPU evaluation; no held-out hyperparameter selection
Runtime and memory

Inference and fit section: 17.6 s.

Metric evaluation section: 0.0481 s.

Device: cpu. Batch size: 4.

These timings describe the recorded sections of this run, not total runtime or a general hardware benchmark. Peak memory is not reported.

Recorded results for this configuration
MetricValueCoverageUncertainty and source
ndcg0.896

dimensionless · higher

184/184

Not reported

Result provenance
spearman-0.146

dimensionless · higher

184/184

Not reported

Result provenance

Uncertainty: Not estimated; one selected evaluation.

Evaluation methods, evidence and reproduction

No execution recipe has been verified for this exact configuration and evaluation. Inspect its methods and original run documentation before attempting reproduction.

Open the protocol's results and comparison checks →

Original execution documentation ↗

Mapping sources and review metadata

Mapping use-case-mapping-rhodopsin-wavelength-rhomax-fixed-probes · revision 1

Initial applicability review: primary sources and separate automated cross-review support this exact protocol, its complete selected comparator group and the stated limits.

Reviewed evidence fingerprint a012ef5867a510fe9861d1a3de019095e38fb8718f0dbd3581ddd1c5b9a51783

What evidence is still missing?

  • A spectral-tuning endpoint is not opsin activation efficiency, expression, photostability, cellular function, general protein fitness or clinical usefulness.
  • Spearman and full-ranking NDCG do not establish wavelength calibration, top-k precision or a prospective experimental hit rate. Constant predictions have undefined Spearman; their NDCG is a control value, not strong predictive evidence.
  • No intervals or seed-variability estimates. The 35M checkpoint was chosen after seeing the 8M outcome; the source describes an exploratory follow-up, not a preregistered family-scale comparison.
  • ESM-2 pretraining overlap is unresolved. A frozen linear probe result does not establish the value of alternative pooling, fine-tuning or every model-family configuration.
  • Recorded timing covers sections of execution and is not a hardware-normalised deployment-cost comparison. Human domain review and external replication are not established.

Contribute evidence or propose a correction

Sources and review

Automated source review · 2026-09-28 · Codex research curation

Primary-source curation and separate automated cross-review checked exact evidence identities, comparator coverage, endpoint relevance and transfer limits. No new execution, human scientific review, independent replication or clinical validation.

Release provenance and downloads

Release 2026-09-28-c7b5ac6d34f2

Use-case input digest a0dd27a5f430ec387d309fd6e8615083250873ce4cff5e882c6fb8458ef80e95

Download questions, collection plans and review metadata (JSON) · Verify release checksums

Question use-case-rhodopsin-wavelength-transfer. Any numerical results on this page come from this release's existing evaluation records.