Efficient PH-ASC Methods

• Efficient PH-ASC is a technique that simulates a limited set of protonation states and reweights their results to construct continuous pH-dependent grand-canonical ensembles.
• The method employs MSM analysis and Fokker–Planck discretization to accurately infer state-to-state transition rates and equilibrium observables with significantly lowered simulation cost.
• Robust clustering via PCCA+ extracts interpretable macrostates and smooth transition rates, enhancing kinetic modeling for peptides and proteins.

Efficient PH-ASC (pH-Dependent Accelerated Sampling & Kinetics) methods are a class of techniques for efficiently computing state-to-state transition rates, equilibrium observables, and kinetic mechanisms in molecular systems with pH-dependent protonation equilibria. These strategies circumvent the high cost of brute-force constant-pH molecular dynamics (MD) by leveraging a minimum set of canonical simulations—one per dominant protonation state—and reweighting their results to reconstruct grand-canonical kinetic and thermodynamic quantities as continuous functions of pH. Recent advancements integrate MSM (Markov State Model) analysis, Fokker–Planck generators, and robust clustering to provide accurate and interpretable pH-dependent kinetics for peptides and proteins with sharply reduced computational effort (Donati et al., 2023).

1. Theoretical Foundations: Grand Canonical Reweighting

Efficient PH-ASC protocols exploit the observation that the full pH-dependent grand-canonical ensemble (GCE) can often be accurately approximated by a small set of canonical ensembles (CEs), each corresponding to a specific protonation microstate (scenario). With $S$ such scenarios, each is simulated under fixed protonation, producing a canonical partition function $Z_n$ and sampled density $\pi_n(x)$. At target pH, these are reweighted by the proton chemical potentials:

$$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$

$$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$

where $N_n$ is the number of protons in scenario $n$, and $\mathcal{Z}(\mathrm{pH})$ is a normalization constant. This framework enables post hoc reweighting to any pH value, provided that enough protonation scenarios are sampled.

2. Kinetic Inference via Markov State Modeling and Fokker–Planck Discretization

The kinetic generator at a given pH, $\mathcal{Q}(\mathrm{pH})$, is discretized on a reduced reaction-coordinate (RC) space partitioned into $K$ cells $Z_n$0, typically using the Square Root Approximation (SqRA):

$Z_n$1

where $Z_n$2 is the area of the interface between cells, $Z_n$3 the center distance, $Z_n$4 the volume of cell $Z_n$5, and $Z_n$6 the effective diffusion coefficient. Calculation of $Z_n$7 uses the grand-canonical formula, interpolating densities from the $Z_n$8 canonical scenarios.

This discretized operator yields a $Z_n$9 rate matrix suited for spectral analysis and coarse-graining, and its construction is efficient for moderate $\pi_n(x)$0 ($\pi_n(x)$1).

3. Coarse-Graining and Rate Extraction: PCCA+ and Macrostates

To extract interpretable transition rates, robust Perron Cluster Cluster Analysis (PCCA+) is used to identify $\pi_n(x)$2 metastable macrostates based on dominant eigenvectors of the kinetic generator. The membership matrix $\pi_n(x)$3 maps cells to macrostates. The coarse-grained rate matrix $\pi_n(x)$4 is computed as:

$\pi_n(x)$5

where the elements $\pi_n(x)$6 quantify transition rates between macrostates $\pi_n(x)$7 and $\pi_n(x)$8 as continuous functions of pH. The approach is robust to the number and character of macrostates and is compatible with high-dimensional RC spaces via mesh-free clustering.

4. Computational Workflow and Scaling

The protocol comprises the following stages:

1. Canonical MD: Simulate $\pi_n(x)$9 protonation scenarios, collect $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$0 in RC space, estimate $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$1.
2. Free Energy and Diffusion Calculation: Obtain $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$2, optionally estimate MSM implied timescales and calibrate $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$3.
3. pH Reweighting and SqRA: For each target pH, compute $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$4, $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$5, and $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$6, then construct $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$7.
4. PCCA+ and Rate Extraction: Identify macrostates and compute $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$8.

The total cost scales as $$w_n(\mathrm{pH}) = \exp\left[\beta N_n \mu(\mathrm{pH})\right],\quad \mu(\mathrm{pH}) = \mu_0 + k_BT \ln 10^{-\mathrm{pH}}$$9, where $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$0 is MD simulation length per scenario, $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$1 the number of RC bins, and $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$2 the number of pH points. Compared to conventional constant-pH MD ($$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$3), the protocol yields a near-linear speedup factor $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$4 for large $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$5 (Donati et al., 2023).

5. Quantitative Performance and Benchmark Results

In the Ala–Asp–Ala model system (S=2: protonated and deprotonated Asp):

• 2 μs MD simulations were performed per scenario.
• Diffusion constants: $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$6 ps$$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$7 (protonated), $$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$8 ps$$\pi(x;\mathrm{pH}) = \frac{1}{\mathcal{Z}(\mathrm{pH})} \sum_{n=1}^S w_n(\mathrm{pH})\,\pi_n(x)$$9 (deprotonated).
• 2D RC space ($N_n$0) covered Ramachandran angles.
• PCCA+ identified $N_n$1 macrostates (β-sheet, $N_n$2, $N_n$3).
• Transition rates $N_n$4 decreased with increasing pH and could be smoothly interpolated over a broad range (10$N_n$5–10$N_n$6 ps$N_n$7).

For larger biomolecules, scalability depends on the dimensionality of the RC space and the number of relevant protonation scenarios. Only scenarios with significant pH weight ($N_n$8) in the target interval need be simulated.

6. Comparison with Alternative and Brute-Force Methods

Conventional constant-pH MD performs separate, full-length simulations at every pH of interest, while efficient PH-ASC performs only $N_n$9 scenario simulations and then analytically interpolates results. This approach preserves rigorous sampling of physical protonation microstates while efficiently mapping pH dependence. Convergence of rates and thermodynamic observables is determined by sampling precision in each scenario, the grid density in RC space ($n$0 discretization error), and robustness of the clustering step. The method has been demonstrated to yield continuous $n$1 with high statistical efficiency and interpretability (Donati et al., 2023).

7. Generalization, Limitations, and Future Directions

Efficient PH-ASC protocols generalize readily to systems with more than two relevant protonation microstates (increasing $n$2), provided their weights are appreciable within the pH range of interest. Incorporation of higher-dimensional RC spaces is possible through advanced clustering (e.g., VAMPnets) and mesh-free spectral clustering (e.g., ISOKANN), subject to tractable $n$3. Incorporating pH-dependent changes in diffusion coefficients is handled by interpolating $n$4 with $n$5 weights. A plausible implication is that for systems with many rarely-populated protonation states, judicious scenario selection is essential for both accuracy and efficiency.

The efficient PH-ASC paradigm provides a unified framework for predicting continuous, interpretable, and physically rigorous pH-dependent kinetics in biomolecular systems using orders of magnitude fewer simulations than traditional constant-pH approaches, with growing applications to peptides, protein folding, and enzyme catalysis (Donati et al., 2023).