STRC Research

MFEP Two-Stage Strategies Table

2 min read

MFEP Two-Stage Strategies Table

P0 data extract — Table 6.1 from 2007-chipot-free-energy-calculations-book §6.2.1, p. 254 (Lu, Woolf, Andricioaei). Verbatim. Catalogues all four directional combinations a two-stage multistage FEP can take when the path passes through one intermediate state M between systems 0 and 1. Free-energy formulas obey ΔA = ΔA_{0→1}; signs follow whichever direction was actually run. Useful any time h01 phase5 or any future alchemical mutation is split into two windows.

Verbatim table

Table 6.1. Example of four staging strategies (perturbation directions) for a two-stage FEP calculation involving the systems 0 and 1, and the intermediate M

Calculation strategyFree energy formula
0 → M → 1ΔA = ΔA_{0→M} + ΔA_{M→1}
0 ← M ← 1ΔA = − ΔA_{M→0} − ΔA_{1→M}
0 → M ← 1ΔA = ΔA_{0→M} − ΔA_{1→M}
0 ← M → 1ΔA = − ΔA_{M→0} + ΔA_{M→1}

The best choice of strategy depends on the problem under study.

— [Chipot & Pohorille 2007, p. 254, Table 6.1]

Decision rule (book §6.3.1)

To pick a strategy, examine the phase-space relationship between Γ₀, Γ_M, Γ*₁ (Phase Space Overlap and FEP Sampling):

  • If Γ₁ ⊆ Γ_M ⊆ Γ*₀ (subset chain) → row 1 (forward only).
  • If Γ₀ and Γ₁ have partial overlap (Fig. 6.1c) → put M in the overlap and use 0 → M ← 1 (row 3, “overlap sampling”): ΔA = ΔA_{0→M} − ΔA_{1→M}.
  • If Γ₀ and Γ₁ have no overlap (Fig. 6.1d) → make Γ*M cover both and use 0 ← M → 1 (row 4, equivalent to umbrella sampling): ΔA = ΔA{M→1} − ΔA_{M→0}.

The book’s bottom line: row 3 (overlap sampling) is the safest default when overlap is uncertain. Row 1 single-direction is only safe when the higher-entropy system is system 0.

How to use this in STRC

  • h01 phase5 alchemical mutation E1659E→E1659A — the wild-type pocket and the mutant pocket are the partial-overlap case. Use the row-3 overlap-sampling staging with M halfway in λ-space; combine with Recipe — Bennett Acceptance Ratio Estimator which is the ML-optimal estimator for that geometry.
  • h26 disulfide cysteine designs — when scoring A1078C, S1080C, S1579C designs against AF3 baseline, dual-topology + overlap sampling reduces the catastrophe risk at λ → 0/1.

Connections

Graph View

Backlinks