A damaging mutation
Identify a mutation with a measured destabilizing effect. Count all tested combinations meeting the damaging-effect cutoff, including those that fail to recover.
Science fair · 2026–2027
The science of
a second chance.
Can a second-site mutation compensate for a destabilizing mutation in the same protein? A computational study of protein folding stability.
Explore the evidence ↗01 / STRUCTURE → RECOVERYPROTEIN MUTATION RESEARCH
RESEARCH IN PROGRESSCurrent focus: Testing candidate robustnessCorrection-consistent candidates · experimental validation pending
01 / The research question
Damage is only
half the story.
A protein’s stability can change when its amino-acid sequence changes. This project asks whether a second mutation can compensate for the first—and whether apparent repair reflects a specific interaction or the partner’s own stabilizing effect.
A second-site partner
Examine the double mutant. Compare its recovery using a physical correction model and an isotonic correction fitted without the held-out observations, always for the same damaging mutation and partner.
A more careful explanation
Check standalone stabilization, partner position, and dependence among related backgrounds before interpreting a repair mechanism.
Recovered stability = partner effect + interactionConceptual decomposition · recovery alone does not isolate epistasis
02 / Explore the evidence
A shared pattern.
A comparison still needed.
In the current analysis, 2K28 E16F, E16W, and E16Y each meet both rescue criteria for the same 15 of 17 tested damaging substitutions. Select a partner to inspect its standalone effect.
2K28 · POSITION E16
15 / 17
Tested backgrounds meeting both correction criteria
E16F standalone effect: −0.132 kcal/mol
Destabilizing on its own
All 17 backgrounds are substitutions at one damaged position: R9. This is not evidence of rescue across 17 positions.
BASELINE NOT YET REPORTED
15 of 17 is a descriptive result.
A “background” here means one damaging amino-acid substitution, such as R9D. The relevant comparison is the fraction of other tested partners that pass on those same substitutions, especially alternatives at E16. Those matched rates and rankings have not yet been computed for this page, so this panel does not establish enrichment.
Interaction effects are not yet reported. Recovery includes both the partner’s standalone effect and its interaction with the damaging mutation. A negative standalone effect rules out simple standalone stabilization as the sole explanation for that partner; it does not by itself quantify a specific interaction.
Matrix: project analysis output supplied by the research team. Standalone effects: the accompanying summary CSV. The two methods share underlying measurements; their agreement is not independent replication.
03 / What the results mean
Promising patterns.
Open explanations.
The purpose of the audit is to test competing explanations, not simply collect significant mutations. These are screening candidates, not experimentally confirmed protein repairs.
Stabilization is not the whole explanation.
Some high-ranking partners are destabilizing on their own. E16F and R27Y have negative standalone effects; D28C and P35W are close to neutral. A strong stabilizing partner cannot explain every candidate.
Compare partners at the same position.
The comparison holds the protein, the damaging substitution, and the partner position fixed. It asks whether one replacement amino acid succeeds more often than alternatives tested at that same position.
Related backgrounds may share information.
The conditional test assumes independent trials across damaged substitutions. Repeated variants at one position and shared measurements make dependence a concern. Multiple-testing correction does not resolve that concern.
One pattern does not identify a mechanism.
Matching aromatic rescue profiles do not show that every E16 substitution works. Compare all tested substitutions at E16, then separate the partner’s own effect from the specific interaction.
04 / Candidate notebook
A shortlist to investigate.
Selected candidates, sorted by the fraction of tested combinations passing both methods. Positive kcal/mol means stabilizing in this project; negative means destabilizing. Protein labels such as 2K28 and 3DKM identify protein domains in the source dataset. These rates describe the tested library, not a universal probability of repair.
| Protein / partner | Both / tested | Standalone effect | Interpretation | Matched baseline |
|---|---|---|---|---|
| 3DKM · N56G | 14 / 14 | +0.438 kcal/mol | Modestly stabilizing | Not yet reported |
| 4G3O · D28C | 14 / 15 | +0.030 kcal/mol | Near neutral | Not yet reported |
| 2K28 · E16F | 15 / 17 | −0.132 kcal/mol | Destabilizing | Not yet reported |
| 1QP2 · P35W | 15 / 18 | +0.052 kcal/mol | Near neutral | Not yet reported |
| 2K5N · R27Y | 12 / 27 | −0.323 kcal/mol | Destabilizing | Not yet reported |
Standalone effects are summary medians. Interaction effects and matched baseline rescue rates are not yet reported on this page. Revised significance remains unverified.
Methods & definitions
What counts as
recovery?
The analysis selects damaging single mutations with a stability effect below −1.0 kcal/mol. The reported screen uses a minimum corrected recovery of 1.0 kcal/mol under each of two correction methods, applied to the same damaging-mutation / partner combination.
The complete event definitions also depend on the original analysis code and its uncertainty checks. Those additional conditions have not been independently reconstructed in this website audit; the two numeric cutoffs alone are not a complete recipe for reproducing the labels. Exact uncertainty rules and correction fits will be documented after verification.
The screen’s conditional p-values assume exchangeable partners within comparison groups and independent trials across damaging substitutions. Shared single-mutant measurements and repeated variants at one position may violate these assumptions. The downloadable audit includes a sensitivity analysis that preserves entire substitution profiles within a damaged-position block.
DATASET CREDIT
Built on published experimental measurements.
Tsuboyama, K. et al. Mega-scale experimental analysis of protein folding stability in biology and design. Nature 620, 434–444 (2023). doi:10.1038/s41586-023-06328-6 ↗
The experimental measurements come from this study. The candidate screen and interpretations shown here are the student project’s secondary analysis, not conclusions attributed to the original authors.
05 / The next experiment
Make every claim
earn its place.
The next stage is a reproducible audit of the original orientation-level data, followed by independent experimental validation of carefully selected candidates.
Downloads are served with this website and require no Drive sign-in. The candidate CSV is a copy of the project’s existing summary; the audit code has not yet been run against the complete original data. This is a fixed research snapshot, not a live connection to Colab.
- Reconstruct each damaging-mutation / partner combination and verify its classification.
- Separate standalone partner effects from interaction effects.
- Compare all tested E16 substitutions on matched backgrounds.
- Test sensitivity to dependence among variants at one damaged position.
- Seek a laboratory partner to plan independent stability measurements of the original protein, each single mutant, and the double mutant. The assay has not yet been selected.
Project team
The people behind
PROTEUS-REPAIR.
James Hong
Adam Shen
Project contact:
jameshongbc@gmail.com