Related Experiment Video
Updated: Jan 8, 2026

Hot Biological Catalysis: Isothermal Titration Calorimetry to Characterize Enzymatic Reactions
Published on: April 4, 2014
Repeatability of Relative Free Energy Calculations in Solution with ANI-2x and MACE-OFF23
Sara Tkaczyk1,2, Thierry Langer1, Marcus Wieder3
1Department of Pharmaceutical Sciences, Pharmaceutical Chemistry Division, University of Vienna, Josef-Holaubek-Platz 2, 1090 Vienna, Austria.
Neural network potentials (NNPs) show promise for alchemical free energy calculations. However, careful evaluation of NNP performance in condensed-phase simulations is crucial for reliable results, as demonstrated by varying performance between ANI-2x and MACE-OFF23(small).
Area of Science:
- Computational Chemistry
- Molecular Modeling
- Physical Chemistry
Background:
- Alchemical free energy calculations are vital for predicting molecular properties.
- Neural network potentials (NNPs) offer a promising, transferable alternative to traditional force fields.
- Assessing NNP suitability for complex simulations, especially in condensed phases, remains an active research area.
Purpose of the Study:
- To investigate the feasibility and challenges of using NNPs for alchemical free energy calculations.
- To evaluate the performance of different NNPs (ANI-2x and MACE-OFF23(small)) in predicting tautomeric states.
- To develop and test a methodology for smooth interpolation between states using energy mixing and dummy atoms.
Main Methods:
- Employed a single-coordinate dual-topology approach for alchemical free energy calculations.
- Utilized energy mixing via selective masking of interactions involving dummy atoms for state interpolation.
- Tested the framework with ANI-2x and MACE-OFF23(small) NNPs for tautomer free energy differences in aqueous solution.
Main Results:
- MACE-OFF23(small) yielded converged free energy results.
- Simulations with ANI-2x exhibited significant variability and convergence issues.
- ANI-2x's performance was hindered by slow water dynamics and solute overstabilization, leading to sampling difficulties.
Conclusions:
- Transferable NNPs are advantageous but require rigorous validation in condensed-phase environments.
- The choice of NNP significantly impacts the reliability of free energy calculations.
- Careful assessment of NNP dynamics and potential for metastable states is essential before application in free energy simulations.
More Related Videos
Related Concept Videos
Calculating Standard Free Energy Changes
An Introduction to Free Energy
Calculating the Equilibrium Constant
For example, gaseous nitrogen dioxide forms dinitrogen tetroxide according to this equation:
Chemical Equilibria: Systematic Approach to Equilibrium Calculations
The first step is to identify all the chemical reactions involved, The...
Gibbs Free Energy
Enthalpy of Solution

