What the evidence shows
RhoMax, which adds structural binding-pocket inputs, had the lowest wavelength error on each of its own wild-type holdout splits.
In separate Rewire runs on the FLIP2 split, two amino-acid composition probes outranked two frozen ESM-2 probes, and both ESM-2 probes fell below a constant control. Averaged over four RhoMax splits, mean absolute error was 10.45 nm for RhoMax, 13.46 nm for RhoMax plus retinal and 14.90 nm for the BLASSO comparator, with standard deviations of 10.9 to 11.7 nm; per-split RhoMax error ran from 7.28 to 16.68 nm. In electronvolts the three methods are closer, and on split 3 all three print the same mean absolute error of 0.045 eV.
On the FLIP2 Rhomax by_wild_type split of 184 held-out sequences, Rewire's runs gave a Spearman correlation of 0.418 for a 40-feature amino-acid composition probe and 0.418 for a 22-feature one, against -0.146 for a frozen ESM-2 8M probe and -0.222 for a 35M probe; NDCG was 0.955, 0.955, 0.896 and 0.907 against 0.921 for a training-mean control, whose Spearman is undefined. The two split definitions are different and their scores must not be pooled. The RhoMax values are reported by its authors with no independent reproduction, and the Rewire runs are single runs with no seed variability, with the 35M checkpoint chosen after the 8M outcome was known.
- You bring
- Complete rhodopsin amino-acid sequences
- Training measurements of peak absorption wavelength and frozen assignments that separate wild-type backgrounds
- You want
- 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.
- Evidence in this release
- 6 comparisons shown · 6 tools · 8 configurations
- Kept up to date
- Evidence last checked 8 October 2026 · re-checked monthly, next due 8 November 2026
Comparisons
Each table is one study or protocol with its own truth set and scoring. Compare tools within a table, not across tables.
RhoMax WT-background holdout split 3
Proxy evidenceAuthor-reportedAbsorption-maximum prediction error in nm and eV on split 3
Sorted by MAE (lower is better). The best value in each column is highlighted.
Scroll sideways to see all 2 columns.
± is a standard deviation.
Caveats and method
- Different split definitions from FLIP2; scores must not be merged.
- WT-group holdout does not prove prospective calibration or functional utility.
- Printed mean row is a summary of split results, not an additional experiment.
- Variant denominators and split hashes remain unextracted.
Protocol: RhoMax WT-background holdout split 3
RhoMax WT-background holdout mean
Proxy evidenceAuthor-reportedAbsorption-maximum prediction error in nm and eV on mean
Sorted by MAE (lower is better). The best value in each column is highlighted.
Scroll sideways to see all 2 columns.
± is a standard deviation.
Caveats and method
- Different split definitions from FLIP2; scores must not be merged.
- WT-group holdout does not prove prospective calibration or functional utility.
- Printed mean row is a summary of split results, not an additional experiment.
- Variant denominators and split hashes remain unextracted.
Protocol: RhoMax WT-background holdout mean
RhoMax WT-background holdout split 1
Proxy evidenceAuthor-reportedAbsorption-maximum prediction error in nm and eV on split 1
Sorted by MAE (lower is better). The best value in each column is highlighted.
Scroll sideways to see all 2 columns.
± is a standard deviation.
Caveats and method
- Different split definitions from FLIP2; scores must not be merged.
- WT-group holdout does not prove prospective calibration or functional utility.
- Printed mean row is a summary of split results, not an additional experiment.
- Variant denominators and split hashes remain unextracted.
Protocol: RhoMax WT-background holdout split 1
RhoMax WT-background holdout split 4
Proxy evidenceAuthor-reportedAbsorption-maximum prediction error in nm and eV on split 4
Sorted by MAE (lower is better). The best value in each column is highlighted.
Scroll sideways to see all 2 columns.
± is a standard deviation.
Caveats and method
- Different split definitions from FLIP2; scores must not be merged.
- WT-group holdout does not prove prospective calibration or functional utility.
- Printed mean row is a summary of split results, not an additional experiment.
- Variant denominators and split hashes remain unextracted.
Protocol: RhoMax WT-background holdout split 4
RhoMax WT-background holdout split 2
Proxy evidenceAuthor-reportedAbsorption-maximum prediction error in nm and eV on split 2
Sorted by MAE (lower is better). The best value in each column is highlighted.
Scroll sideways to see all 2 columns.
± is a standard deviation.
Caveats and method
- Different split definitions from FLIP2; scores must not be merged.
- WT-group holdout does not prove prospective calibration or functional utility.
- Printed mean row is a summary of split results, not an additional experiment.
- Variant denominators and split hashes remain unextracted.
Protocol: RhoMax WT-background holdout split 2
FLIP2 Rhomax by_wild_type
Proxy evidenceRewire runSpearman rank association and full-ranking NDCG for absorption-wavelength predictions on 184 held-out Rhomax sequences.
Sorted by NDCG (higher is better). The best value in each column is highlighted.
Scroll sideways to see all 2 columns.
Caveats and method
- 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.
Protocol: FLIP2 Rhomax by_wild_type
Not covered yet
- 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.
- 30 September 2026 audit: Original RhoMax splits group 75 WT backgrounds (65 train/10 test per split) and are not the single FLIP2 by_wild_type partition. Do not pool their scores.
- 30 September 2026 audit: Reported standard deviations describe error dispersion; printed aggregate is a summary of four splits, not an independent experiment or pooled uncertainty interval.
- 30 September 2026 audit: New-background prospective calibration, target-specific candidate hit rates, expression, activation and photostability remain unestablished.
- 30 September 2026 audit: RhoMax Tables2–3 are feature/attention ablations; they were inspected but intentionally not added to this model-selection comparison.
Contribute evidence or propose a correction
Question, sources and review
Who this is for, setting and exclusions
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
- 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.
- Research setting
- Research only. Absorption-wavelength ranking does not establish suitability for an optogenetic intervention, patient care or treatment.
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
Sources, review and downloads
Automated source review · 2026-09-30 · Codex source curation, independent worker cross-review and root integration review
Source-backed literature curation with independent automated transcription and scope checks. No new model execution, independent experimental replication, qualified human scientific review or clinical validation.
Release 2026-11-02-a7f03c9e55a5 · question use-case-rhodopsin-wavelength-transfer
Use-case input digest 43a80aef5473f1a49c013b6d46b720b850ff5a0d18dcbb741dcdeed1ed082db1
Download questions and review metadata (JSON) (gzip) · Verify release checksums (gzip)