| name | design-thermostable-mutations |
| description | Decision-grade enzyme/protein mutation design for thermostability with bioactivity-preserving constraints enforced by default, plus structure-aware and consensus-ranking workflows. |
Skill: Design Thermostable Mutations
Use When
Use this skill when users need mutation design or triage for proteins/enzymes and care about both:
- improved thermostability/heat tolerance
- preserved catalytic activity/bioactivity
Typical requests:
- "Which positions should I mutate first?"
- "Design a stability-focused mutation library without breaking active site chemistry"
- "Rank these candidate mutations by stability evidence"
- "Build a decision-grade shortlist for wet-lab validation"
- "Predict the stability impact of a specific variant (e.g., a clinical or engineered substitution)"
Decision-Grade Guardrails (Default)
These are enforced by default unless explicitly overridden:
- Bioactivity constraints must be supplied for design decisions:
- active-site/catalytic/cofactor/disulfide/blocklist positions
- Functional/protected positions are blocked from routine mutagenesis.
- Non-conservative substitutions are rejected in conservative mode.
- User-facing mutation notation is 1-based and WT residue-validated.
Inputs
- Required:
- WT sequence (
--sequence or --fasta)
- Required for decision-grade design mode:
- functional constraints (at least one: active/cofactor/disulfide/blocklist)
- Optional:
- structure (
.pdb/.cif/.mmcif)
- mutation candidates (
A123V format)
- assay context (pH, temperature, ligand/cofactor state)
Workflow
Choose the smallest workflow that still supports decision-grade confidence.
Path A: Fast Heuristic Design (single-script)
Use when structure and external predictors are unavailable.
Script:
scripts/design_thermostable_mutations.py
Capabilities:
- strict sequence/mutation validation
- protected-residue enforcement (default)
- conservative-substitution enforcement (default)
- heuristic DDG + local aggregation-risk filtering
- ranked candidate output with rejection reasons
Path B: Structure-Aware Consensus (recommended)
Use when structure is available and shortlist quality matters.
- One-command orchestrator (recommended entrypoint):
scripts/run_decision_pipeline.py
- Discover candidate sites with structural context and functional guardrails:
scripts/discover_candidate_sites.py
- Generate focused single/multi-site library:
scripts/generate_mutation_library.py
- Resolve sequence-to-structure numbering safely:
scripts/structure_residue_mapper.py
- Run structure-based scoring (FoldX wrappers):
scripts/run_foldx_batch.py
scripts/run_foldx_chunked.py (large libraries)
- Aggregate multiple metrics with direction-aware consensus:
scripts/consensus_stability_rank.py
- Produce decision-ready report using:
- ranked consensus
- support fraction across tools
- functional risk flags
- assumptions and escalation notes
- Optional AI-assisted route for structure generation/summary:
scripts/run_colabfold_batch.py
Example:
python3 scripts/run_decision_pipeline.py \
--sequence-fasta wt.fasta \
--structure model.pdb \
--active-site 57,102,195 \
--cofactor-sites 64,66 \
--blocklist-sites 1-5 \
--run-foldx \
--run-consensus
By default, the pipeline:
- requires functional constraints for decision-grade discovery
- blocks protected sites during library generation
- auto-derives protected sites from active/cofactor/disulfide/blocklist when not explicitly set
Output Contract
Always include:
- Baseline stability summary
- Functional-constraint enforcement status
- Ranked candidates with rationale
- Rejection summary (rule-level counts)
- Functional risk flags (active/cofactor/disulfide/interface/conservation)
- Confidence and escalation section
For consensus workflows, also include:
- per-tool metric table
- consensus score and support fraction
- metric directionality and normalization choices
References
references/mutation-design-workflow.md
references/mutation-design-principles.md
references/stability-prediction-playbook.md
references/use-case-playbook.md
references/report-template.md
references/troubleshooting.md
references/literature-evidence-2022-2026.md
Evaluation and Bundle Checks (Optional)
evals.json
scripts/run_skill_evals.py
scripts/validate_skill_bundle.py
Install
From repository root:
./scripts/install_skill.sh design-thermostable-mutations
./scripts/install_skill.sh design-thermostable-mutations --tool claude
./scripts/install_skill.sh design-thermostable-mutations --dest ~/.my-tool/skills
Related Skills
- predict-protein-heat-stability: Use this first to assess baseline thermostability of a sequence or compare variant heat stability before launching mutation design. It provides Tm class prediction, composition features, and variant ΔTm analysis without requiring mutation design intent.
Limits
- Heuristic scores are triage evidence, not biochemical truth.
- Predictions must be experimentally validated (Tm/T50/activity).
- If functional constraints are incomplete, activity-preservation confidence drops.
- Path A heuristic DDG model (
estimate_ddg_simple) models local hydrophobic core packing; it underestimates surface-exposed positions where Arg/Lys enrichment and salt bridges drive thermostability. Use Path B for surface sites.
- Common failure cases:
- structure numbering mismatches that are not validated before scoring
- incomplete active-site/cofactor/disulfide/blocklist constraints
- overlarge mutation libraries that exceed practical scoring budgets