Fast Fourier Transform Enables Automated Parametrization of Complex Dihedral Potentials in All-Atom and Coarse-Grained Force Fields

Torsional parametrization remains one of the most persistent challenges and weaknesses of modern force fields, particularly for dihedrals whose asymmetry and multimodality evade traditional Fourier or Ryckaert–Bellemans treatments [Shirts, M. R.; Mobley, D. L. In Biomolecular Simulations, Springer, 2013; Vol. 9, pp 71–120 andBowman, J. M. Chem. Rev. 2017, 117, 10034–10072]. Here, we introduce a general and fully automated methodology for dihedral parametrization in both all-atom (AA) and coarse-grained (CG) models based on the Fast Fourier Transform (FFT). This Fourier-analysis framework provides a systematic and unbiased route to reconstruct torsional energy profiles of arbitrary complexity, including nonsymmetric and multimodal shapes that have remained inaccessible to existing parametrization tools. When combined with iterative refinement via QM–MM energy matching in AA models and Iterative Boltzmann Inversion in CG models, our FFT approach yields torsional potentials that quantitatively reproduce reference energy landscapes across a wide variety of chemical environments. We apply this methodology to different molecules within an AA framework, obtaining consistently improved agreement with QM reference profiles. In the CG regime, our method is demonstrated on two systems, the MS-Z molecular switch and the Aβ42 peptide, yielding transferable torsional potentials that enable accurate modeling of their conformational behavior. Overall, this work demonstrates that a FFT-based torsional parametrization is a robust and general strategy for developing next generation force fields.