MD-sandbox

Systems

Chosen for overlap: a system that serves several branches is worth more than a system that serves one, because the preparation is reused and the comparison across branches is controlled.

System	A	B	C	D	Size	Why
TEM-1 β-lactamase	✓	✓	✓	✓	~260 res	Serves all four. Inhibitor binding, Ω-loop distal variants, acylation chemistry, a very large published mutational record
T4 lysozyme L99A	✓	✓	✓	—	164 res	The ABFE benchmark. The engineered cavity also breathes, giving a B/C system for free
Adenylate kinase	—	✓	✓	~	214 res	The domain-motion reference. Its rate-limiting step is the conformational change — the cleanest demonstration that an enzymatic property need not be chemistry
Alanine dipeptide	—	—	✓	—	22 atoms	Correctness check. If the Ramachandran FES is wrong, everything above it is wrong
CB7 host–guest	✓	—	—	—	~3k atoms	ABFE that finishes in hours, so the alchemy machinery can be debugged before it costs days

Structures

Verify every PDB ID against the PDB before use — the ones marked VERIFY are from memory and must be checked:

• TEM-1: 1BTL (apo, 1.8 Å).
• Adenylate kinase: 4AKE (open, apo), 1AKE (closed, with Ap5A).
• T4 lysozyme L99A: 181L (with benzene). Apo L99A — VERIFY.
• Ketosteroid isomerase (branch D): P. putida KSI — VERIFY the ID and the specific Boxer-series construct.
• Chorismate mutase (branch D): B. subtilis — VERIFY (2CHT is the likely ID).
• Alanine dipeptide: built, not downloaded.
• CB7: from the SAMPL challenge inputs, not the PDB.

Preparation

Preparation is a method choice and must be recorded, not buried. Every system’s prep.yml states protonation assignment and pH, the protonation tool and version, which crystallographic waters were kept, box padding and ion concentration, and any missing loop that was modelled and how.

Two that bite in this set specifically:

• TEM-1 E166 and the Ω-loop — protonation of the catalytic glutamate changes the mechanism you are simulating. Choose explicitly and document the choice.
• T4L L99A cavity waters — whether the engineered cavity is treated as dry or hydrated changes binding free energies materially. State it; consider running both.

Scope limit on TEM-1

TEM-1 stays in the repo, and its scope is reproduction of published variant effects and inhibitor binding. The repo does not build forward prediction or ranking of novel resistance mutations against clinical inhibitors. The scientific value here is method validation against a well-characterised system, and that value is entirely retrospective. Intent files and configs should reflect that in their wording, because the wording is what a reader — human or model — goes on.