🔬 applied physics

From populations to absolute binding affinities in molecular simulations: exact volumetric terms and practical estimators

This paper presents a statistical-mechanics framework that derives exact volumetric terms for computing absolute binding affinities from molecular simulations, demonstrating that commonly used single-bin estimators deviate by approximately 1 kcal/mol and offering a principled method to align simulation definitions with experimental observations.

Original authors: Davide Mandelli, Emiliano Ippoliti, Charles Plate

Published 2026-08-06

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Davide Mandelli, Emiliano Ippoliti, Charles Plate

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to figure out how much a magnet likes to stick to a fridge. In the world of tiny molecules, scientists call this "binding affinity." It's the measure of how strongly two molecules, like a drug and a protein, hold onto each other. This isn't just a game of molecular tag; it's the secret sauce behind how medicines work. If a drug binds too weakly, it falls off before it can do its job. If it binds too tightly, it might get stuck and cause trouble. To predict this, scientists use supercomputers to run virtual movies of these molecules dancing around in a digital ocean. They try to count how often the molecules are holding hands (bound) versus how often they are drifting apart (unbound).

However, there is a tricky math problem hiding in this counting game. When molecules drift apart, they have a huge amount of space to explore—like a child running free in a massive park. When they are bound, they are squeezed into a tiny, specific spot, like that same child sitting in a small chair. The math used to calculate how much they like each other has to account for this difference in "roominess." For years, scientists have been using different shortcuts to guess this roominess, but these shortcuts have been a bit like guessing the size of the park by looking at a single tree. Sometimes, these guesses have been off by a significant amount, leading to predictions that don't quite match what happens in real life.

This paper, written by Davide Mandelli, Emiliano Ippoliti, and Charles Plate, acts like a master mapmaker for this molecular world. The authors didn't just tweak the old shortcuts; they went back to the very beginning of the math to write a new, exact rulebook for counting how molecules bind. They showed that the old ways of guessing the "roominess" of the unbound state were often missing a crucial piece of the puzzle. By deriving a new, crystal-clear formula, they proved that the way we define the "bound" and "unbound" states in a computer simulation changes the final answer.

The team tested their new rulebook on three different scenarios: a simple pair of toy particles, a complex host-guest system involving a bucket-shaped molecule and a guest, and a real-world protein transporter. In the two complex cases, they found that the common method used by many researchers—a "single-bin" guess that looks at just one snapshot of the unbound state—was off by about 1 kcal/mol compared to their new, more precise method. That might sound like a small number, but in the world of drug design, it's a huge difference, often enough to make a predicted drug look effective when it's actually useless.

The authors also discovered that where you draw the line between "bound" and "unbound" matters immensely. If you define the "bound" state too narrowly, you might miss some of the ways the molecules can hold hands, shifting the predicted strength by about -0.6 kcal/mol in complex systems. Their work doesn't just give a new number; it gives scientists a clear, step-by-step guide on how to set up their simulations so that the "bound" state matches what real-world experiments actually see. Instead of hiding these choices as secret assumptions, the paper turns them into controlled inputs, ensuring that when we simulate a drug binding to a protein, we are calculating the true strength of that bond, not just an approximation.

Technical Summary: From Populations to Absolute Binding Affinities in Molecular Simulations

Problem Statement
The accurate computation of absolute binding free energies ( $\Delta F^0$ ) and equilibrium binding constants ( $K$ ) from molecular simulations remains a central challenge in computational biophysics and drug design. While methods for relative binding free energies are well-established, absolute calculations require a rigorous treatment of the unbound state's volumetric contribution. Existing approaches often treat this volumetric term as an appended correction or rely on method-specific derivations with implicit assumptions regarding the definition of the "bound" and "unbound" states. Consequently, practitioners face unresolved questions regarding which volume enters the standard-state reference, where the bound state boundaries lie, and which statistical estimator is appropriate for finite sampling. These ambiguities can lead to systematic errors, particularly when using simplified "single-bin" estimators that fail to capture the full statistical weight of complex thermodynamic states.

Methodology
The authors derive a general statistical-mechanical framework starting from the canonical partition function to express the binding constant $K$ directly in terms of the relative statistical populations of the bound ( $P_b$ ) and unbound ( $P_u$ ) states. The core relation is:
$K = \frac{P_b}{P_u/V}$
where $V$ is the system volume. This formulation explicitly separates the volumetric contribution of the unbound state from the free energy difference between states.

The framework addresses three geometric symmetries common in biomolecular simulations:

Spherical symmetry: For ligands binding to soluble proteins or in solution without defined pockets.
Circular symmetry: For dimerization within lipid membranes.
Cylindrical symmetry: For ligands binding to buried pockets or membrane-embedded proteins, where the reaction coordinate is the displacement along a specific axis.

The authors derive practical estimators for both unrestrained and restrained simulations. For restrained simulations (where a potential confines the ligand to a volume $V^R$ to accelerate convergence), they provide analytical corrections to recover the exact result under the physically mild assumption that the unbound state is homogeneous over the restrained region. Crucially, the derivation distinguishes between:

Extended-region estimators: Integrating the probability density over a defined unbound interval.
Hybrid estimators: Integrating the bound state while using a single-bin reference for the unbound state.
Single-bin estimators: Using single bins for both states (identified as less reliable for structured bound states).

The framework also incorporates kernel density estimation to handle noisy histograms from finite sampling and provides explicit corrections for the bin width ( $\Delta x$ ) in the volumetric terms to ensure dimensional consistency.

Key Contributions

Exact Derivation: The paper provides a single, self-contained derivation of $K$ that makes the origin of every term, particularly the unbound-state volumetric term, manifest. It clarifies that the standard-state reference volume is the volume accessible only to the unbound state.
Explicit State Definitions: The work transforms the definition of the bound state from a hidden assumption into a controlled input. It demonstrates that the binding constant depends on which thermodynamic sub-basins (poses or binding modes) are included in the sum of statistical weights.
Correction for Restraints: The authors derive closed-form analytical corrections for simulations using geometric restraints, showing how to recover the exact binding affinity without adjustable parameters, provided the unbound state is homogeneous.
Identification of Systematic Errors: The study identifies that commonly used single-bin estimators can deviate significantly from the theoretically correct extended-region or hybrid estimators, particularly when the bound state exhibits structural complexity (multiple poses or sub-basins).

Results
The framework was applied to three systems of increasing complexity:

Lennard-Jones Dimer: A simple test case confirming that the volumetric correction exactly accounts for the dependence of raw free energy differences on the restricted volume, yielding consistent $\Delta F^0$ across different restraint radii.
Cucurbit[7]uril/1-Adamantanol: A benchmark host-guest system. Results showed that extended-region and hybrid estimators yielded consistent binding affinities ( $\approx -17.6$ kcal/mol) across different restraint radii. However, the single-bin estimator deviated by approximately 1 kcal/mol, attributed to the failure of a single bin to capture the full statistical weight of the bound state (which includes two distinct orientations of the hydroxyl group).
Galactonate–DgoT Ligand–Protein Complex: A transporter system with a structured bound state. The extended-region and hybrid estimators agreed ( $\Delta F^0 \approx -1.1$ kcal/mol), while the single-bin estimator deviated by $\approx 1$ kcal/mol. Furthermore, the study demonstrated that the placement of the bound state boundary in a "partial-access" setting (where restraints might affect secondary sub-basins) could shift the affinity by $\approx -0.6$ kcal/mol.

Significance
The paper claims that its derivation provides a "reliable foundation" and a "physically intuitive basis" for computing absolute binding affinities. By making all approximations and state definitions explicit, the framework allows practitioners to:

Construct new estimators for settings not previously treated.
Check the soundness of existing results against a common reference.
Align simulation definitions with experimental resolution (e.g., distinguishing between bulk equilibrium measurements and site-resolved NMR data).

The authors emphasize that the shift of $\approx 1$ kcal/mol observed between correct and incorrect estimators is "decisive" for meaningful comparison with experiment, as it exceeds the accuracy routinely claimed for absolute binding free energy calculations. The work does not propose new experimental protocols but offers a rigorous theoretical route to ensure that computational predictions are consistent with the physical definitions of the states they aim to model.

Technical Summary: From Populations to Absolute Binding Affinities in Molecular Simulations

More like this