structural-strategy

name

structural-strategy

description

Decision strategies for macromolecular structure building into cryo-EM maps. Use when deciding WHAT to do, in WHAT ORDER, and WHY — not how to run specific tools (that's chimerax/isolde/phenix/ccp4/emerald skills). Covers fitting, model building, refinement, validation, and special cases. Load when planning a structure-building task, choosing between approaches, or troubleshooting a stuck pipeline step.

Structural Biology Strategy

Distilled decision-making knowledge for cryo-EM model building. This skill answers "what should I do next?" and "which approach works here?" — not "how do I run this command."

How to use this skill

1. Read this file first — it has the decision trees and quick rules
2. Load a references/ file only when you need detailed strategy for a specific topic
3. After completing a project step, distill new strategies back into the appropriate references/ file

Quick Decision Trees

"I have a predicted model and a cryo-EM map. Now what?"

1. What resolution?
   ├─ <2.5Å → aggressive: full pipeline, expect near-atomic features
   ├─ 2.5-3.5Å → standard: pipeline works well, cautious with small features  
   └─ >3.5Å → conservative: ISOLDE essential, skip small-feature modeling

2. How many chains/domains?
   ├─ Single domain → rigid-body fit, skip domain segmentation
   ├─ Multi-domain, single chain → Merizo segmentation, per-domain fitting
   └─ Multi-chain complex → Merizo on each chain, careful with interfaces

3. Is it an AlphaFold/predicted model?
   ├─ Yes → expect register errors, domain orientations wrong, loops unreliable
   │        Trim low-pLDDT regions BEFORE fitting
   └─ No (experimental) → less trimming needed, but check for crystal packing artifacts

"The fit looks wrong / metrics aren't improving"

1. Check map-model agreement first
   ├─ Global CC < 0.3 → fitting failed. Re-run with more rotation samples or manual placement
   ├─ CC 0.3-0.5 → partial fit. Some domains may be misplaced → per-domain fitting
   └─ CC > 0.5 → reasonable. Issues are local, not global

2. Identify the problem region
   ├─ Backbone register error → ISOLDE (physics-based correction)
   ├─ Sidechain wrong → rotamer fix protocol (swapaa → ISOLDE → gentle Phenix)
   ├─ Loop in wrong place → trim + rebuild, or ISOLDE with restraints released
   ├─ Domain orientation wrong → re-fit that domain specifically
   └─ Density is just bad → accept it, lower expectations for that region

3. Resolution-dependent expectations
   ├─ Rotamer outliers at >3Å → many are real, don't over-fix
   ├─ Waters at >2.5Å → don't model them
   ├─ Metal coordination at >3Å → geometry restraints essential (can't rely on density alone)
   └─ DNA/RNA at >3.5Å → backbone trace only, don't trust base orientations

"Should I use ISOLDE or Coot for this fix?"

ISOLDE when:
  - Large backbone movements needed (>2Å shift)
  - Register errors (entire segment shifted)
  - Flexible regions that need physics-based relaxation
  - Low resolution (<3.5Å) where manual building is unreliable
  - Multiple simultaneous problems in one region

Coot when:
  - Single residue fixes (flip, rotamer change)
  - High resolution (<2.5Å) where density is unambiguous
  - Quick checks (real-space correlation per residue)
  - Experienced user who prefers manual control

Neither — just Phenix when:
  - Small geometry fixes (bond lengths, angles)
  - B-factor refinement needed
  - Final polishing after ISOLDE

"Which refinement engine should I use?"

Cryo-EM workflow:
  ├─ Intermediate cleanup (between Coot/ISOLDE rounds)
  │   → Phenix real_space_refine
  │     (rotamer fitting, Ramachandran restraints, local geometry)
  │
  ├─ Final refinement for deposition
  │   → Servalcat refine_spa
  │     (half-map input, FSC weighting, B-factors, better statistics)
  │
  ├─ Geometry-only cleanup (no map needed)
  │   → Servalcat refine_geom
  │
  └─ B-factor refinement
      → Servalcat refine_spa (more robust than Phenix for this)

X-ray workflow:
  ├─ Standard refinement → Refmac5 or Phenix refine
  └─ Fine-grained control (TLS, twins, NCS) → Refmac5

Recommended cryo-EM pipeline:
  Coot edit → Phenix RSR (iterate) → Servalcat refine_spa (final) → validate → deposit

Execution note: this skill chooses the engine. When the chosen command is Refmac5, Servalcat, refmacat, AceDRG, or another CCP4 binary, load the ccp4 skill to run it. When the chosen command is Rosetta EMERALD / GALigandDock with density, load the emerald skill to run it.

"Do I need gemmi for this?"

gemmi is a general-purpose prep/inspection tool. Use it when:
  - Converting between mmCIF ↔ PDB formats
  - Inspecting/repairing cell/symmetry/header info
  - Slicing or querying maps
  - Quick residue-level queries on a structure
  - Any prep step that doesn't need a full refinement engine

No skill needed — CLI (gemmi --help) and Python API (import gemmi) are self-explanatory.
Available at: /opt/homebrew/bin/gemmi (Homebrew) + ~/ccp4-9/bin/gemmi (CCP4)

Core Principles

1. Always start with the best possible rigid-body fit. Everything downstream depends on it. More rotation samples = better. Center model on map first.

2. Work from large to small. Global fit → domain fit → ISOLDE → sidechain → waters/ions. Never start fixing sidechains when the backbone is wrong.

3. Preserve completeness. Never extract domains and reassemble — you'll lose linkers. Fit copies, get transforms, apply back to the complete model.

4. Resolution sets expectations. At 3.3Å you won't get perfect rotamers. At 4Å you're doing backbone trace. Don't fight the data.

5. Iterate, don't optimize in one shot. Phenix round 1 → check → adjust restraints → round 2. ISOLDE 5min → check → ISOLDE 5min. Small steps.

6. Validate continuously. Don't wait until the end. Check CC, MolProbity, Ramachandran after every major step. Catch problems early.

7. Know when to stop. MolProbity < 2.0, Rama favored > 97%, clashscore < 10, CC_mask reasonable for resolution. Diminishing returns are real.

8. Trust physics over statistics at low resolution. ISOLDE's forcefield prevents impossible geometry. Phenix can over-fit at low resolution.

9. Metals need explicit restraints. Phenix won't get coordination geometry right without .edits files. Literature distances are your friend.

10. Save everything, overwrite nothing. model_N_description.cif. You will want to go back.

Reference Files

File	When to load	Content
references/fitting.md	Planning fitting strategy	Rigid-body, domain fitting, flexible fitting, convergence criteria
references/model-building.md	Building/modifying model components	DNA/RNA, ligands, metals, gap filling, loop building
references/refinement.md	Planning refinement rounds	Phenix strategy, restraint design, iterative approaches
references/validation.md	Assessing/improving quality	Metrics interpretation, fix protocols, when to stop
references/special-cases.md	Non-standard situations	Membrane proteins, glycans, heterogeneity, low resolution