# Reference protocol and permeability yardsticks

The MD pipeline in this folder is a deliberate copy of one published protocol,
with the numbers it should be read against collected here.

## Sources

- **"Linker-Dependent Folding Rationalizes PROTAC Cell Permeability,"
  *J. Med. Chem.* 2022 (PMC9574858).** The MD paper, and the template for
  `protac_md.py` and `analyze_conformations.py`. Three CRBN/BRD4 PROTACs with
  different linkers, PEG-like against alkyl.
- **Atilaw et al., "Solution Conformations Shed Light on PROTAC Cell
  Permeability," *ACS Med. Chem. Lett.* 2021, 12, 107-114.** NMR plus Monte
  Carlo conformational search and NAMFIS, not MD, on one VHL PROTAC (ERK5). It
  is the source of the folded-versus-extended framing and of the Rg cutoff
  below, nothing else.

## Protocol

| Setting | Reference choice | Here |
|---|---|---|
| Engine / ligand force field | Amber 18, GAFF, geometry pre-optimized in Gaussian | OpenMM 8, OpenFF Sage 2.2 with AM1-BCC charges |
| Solvent | explicit chloroform, ~3000 molecules, truncated octahedron | explicit chloroform via packmol, plus water as a polar reference |
| Production | 100 ns after equilibration | 50-100 ns, `--prod-ns` |
| Replicas | 3 | 3, independent velocity seeds |
| Ensemble | 300 K, PBC, ~10,000 snapshots per trajectory | 300 K / 1 bar NPT, 10 ps per frame |
| Convergence | RMSD, converged within 5-10 ns | RMSD to the first frame plus replica-mean spread |
| Analyses | Rg (cpptraj), 3D PSA (PyMOL), intramolecular H-bonds, PMI shape, PCA and K-means into 5 clusters | Rg, 3D PSA (mdtraj SASA over N, O and polar H), H-bonds, PCA and K-means |

Chloroform stands in for the membrane interior: its dielectric constant of 4.8
is close to what a molecule sees while crossing a bilayer.

## Yardsticks

- **Rg below 7 A** marks the folded, passively permeable regime. This cutoff is
  from the 2021 NMR paper, on a single VHL PROTAC, and it does not separate the
  compounds in the 2022 MD paper.
- **Mean Rg alone does not discriminate.** In the 2022 MD paper:

  | compound | P_passive (nm/s) | dominant Rg | compact local minimum |
  |---|---|---|---|
  | PROTAC 1, most permeable | 30 ± 1.5 | just over 8 A | 5.6 A |
  | PROTAC 2 | 11 ± 1.7 | just over 7 A | 5.5 A |
  | PROTAC 3, least permeable | 6 ± 1.4 | 7-9.5 A, extended | none |

  Their most permeable compound sits mainly just above 8 A, overlapping their
  worst. What separates the two is whether a compact minimum near 5.5 A is
  accessible at all: PROTACs 1 and 2 have one, PROTAC 3 does not. Report the Rg
  distribution and the compact-state population, which is what `rg_basins.py`
  computes, rather than the mean. Absolute Rg is also not strictly transferable
  across chemical series; theirs is CRBN/BRD4, the runs here are VHL/EGFR.
- **3D PSA** rises as permeability falls: 190-265 A^2 for the two more permeable
  compounds against 290-330 A^2 for the poorly permeable one. The rough cutoff
  for low-to-moderate permeability is 200-210 A^2, from the 2021 paper. Absolute
  values depend on the polar-atom selection; `validate_psa.py` and
  `psa_method_probe.py` pin down where this convention sits relative to theirs.
- Permeability correlates inversely with 3D PSA and with intramolecular H-bond
  count.