Lambda-ABF: Simplified, Portable, Accurate and Cost-effective Alchemical Free Energy Computations

Louis Lagardère,Lise Maurin,Olivier Adjoua,Krystel El Hage,Pierre Monmarché,Jean-Philip Piquemal,Jérôme Hénin

2024-05-14

Abstract:We introduce an efficient and robust method to compute alchemical free energy differences, resulting from the application of multiple walker Adaptive Biasing Force (ABF) in conjunction with strongly damped Langevin $\lambda$-dynamics. Unbiased alchemical free energy surfaces are naturally recovered by Thermodynamic Integration (TI). No manual optimization of the $\lambda$ schedule is required as the sampling of the $\lambda$ variable is continuous and converges towards a uniform distribution. Free diffusion of $\lambda$ improves orthogonal relaxation compared to fixed $\lambda$ methods such as standard TI or Free Energy Perturbation (FEP). Furthermore, the multiple walker strategy provides coverage of orthogonal space in a generic way with minimal user input and negligible computational overhead. Of practical importance, no adiabatic decoupling between the alchemical and Cartesian degrees of freedom is assumed, ensuring unbiased estimates for a wide envelope of numerical parameters. We present two high-performance implementations of the method in production molecular dynamics engines, namely NAMD and Tinker-HP, through coupling with the Colvars open source library. These interfaces enable the combination of the rich feature sets of those packages. We demonstrate the correctness and efficiency of the approach on several real-world cases: from solvation free energies up to ligand-receptor binding (using a recently proposed binding restraint scheme) with both fixed-charge and polarizable models. We find that, for a chosen accuracy, the computational cost is strongly reduced compared to state-of-the-art fixed-lambda methods and that results within 1~kcal/mol of experimental value are recovered for the most complex system. The implementation is publicly available and readily usable by practitioners of current alchemical methods.

Chemical Physics

What problem does this paper attempt to address?

This paper aims to solve the problem of calculating free - energy differences in the fields of computational chemistry and biology. In particular, by introducing a new method - lambda - ABF (λ - ABF), it calculates alchemical free - energy differences in a simplified, portable, accurate and cost - effective way. Specifically, the paper attempts to solve the following problems:

1. **Limitations of traditional methods**: Existing alchemical free - energy calculation methods usually require manual optimization of the λ - schedule (i.e., the transition path between different states), and post - processing steps are required when calculating free - energy differences. These processes are not only complex but also time - consuming, limiting the use by non - expert users.

2. **Improving sampling efficiency**: The traditional fixed - λ method has difficulties in sampling slow orthogonal degrees of freedom (such as solvent reorganization or binding - site relaxation), which limits the efficiency and accuracy of the calculation. The method proposed in the paper improves the sampling efficiency of these orthogonal degrees of freedom by continuously sampling the λ variable and using the multiple - walker strategy.

3. **Simplifying user operations**: The method proposed in the paper enables users to obtain free - energy estimates more simply by using the Adaptive Biasing Force (ABF) scheme, without the need for complex parameter adjustments or post - processing steps.

4. **Expanding the application range**: The paper shows the application of this method in biomolecular systems, including the calculation of ligand - receptor binding, whether using a fixed - charge model or a polarizable model. This indicates that the method has wide applicability and robustness.

### Method features

- **λ - propagation**: Under strong - damping Langevin dynamics, the λ variable has a large mass, enabling it to transition smoothly between different λ values.
- **Adaptive biasing**: Through the Adaptive Biasing Force (ABF) scheme, the bias of the λ variable is dynamically adjusted to overcome free - energy barriers.
- **Multiple - walker strategy**: Multiple independent walkers are used for simulation, and the current average - force estimates are regularly shared to update the local bias, thereby improving the sampling efficiency.
- **Free - energy calculation**: The free - energy difference is calculated by the Thermodynamic Integration (TI) estimator.

### Theoretical basis

- **λ - dynamics principle**: λ is regarded as a dynamic variable, associated with a fictitious mass \( m_\lambda \), and an extended Hamiltonian \( H(q, p; \lambda, p_\lambda) \) is defined. The dynamics of the extended system is simulated by Langevin dynamics, enabling λ to be continuously sampled in the [0, 1] interval.
- **Parameter selection**: Appropriate mass \( m_\lambda \) and friction coefficient \( \gamma_\lambda \) are selected to ensure effective sampling in the λ - space and numerical stability.
- **Adaptive biasing force**: The free - energy gradient of λ is adaptively calculated by the ABF algorithm and applied to the simulation system, making the sampling of λ tend to be uniformly distributed, thereby promoting the crossing of free - energy barriers.

### Application examples

- **Simple solvation free - energy**: The application of this method in the calculation of simple solvation free - energy is shown.
- **Protein - ligand binding free - energy**: The absolute free - energy of protein - ligand binding is calculated using a fixed - charge model (such as CHARMM) and a polarizable model (such as AMOEBA), verifying the robustness and efficiency of the method.

In conclusion, by introducing the lambda - ABF method, this paper provides a simplified, efficient and accurate alchemical free - energy calculation method, which is suitable for free - energy calculations in various biomolecular systems.