The computational study of conformational transitions in nucleic acids still faces many challenges. For example, in the case of single stranded RNA tetranucleotides, agreement between simulations and experiments is not satisfactory due to inaccuracies in the force fields commonly used in molecular dynamics. Improvement of force fields is however hindered by the diiculties of decoupling those errors from the statistical errors caused by insuicient sampling. We here tackle both problems by introducing a novel enhancing sampling method and using experimental data to improve RNA force fields. In this novel method, concurrent well-tempered metadynamics are integrated in a Hamiltonian replica-exchange scheme. The ladder of replicas is built with different strength of the bias potential exploiting the tunability of well-tempered metadynamics. Using this method, free-energy barriers associated to individual collective variables are significantly reduced compared with simple force-field scaling. The introduced methodology is flexible and allows adaptive bias potentials to be self-consistently constructed for a large number of simple collective variables, such as distances and dihedral angles. Additionally, a modified metadynamics algorithm is used to calculate correcting potentials designed to enforce distributions of backbone torsion angles taken from experimental structures. Replica-exchange simulations of tetranucleotides including these correcting potentials show significantly better agreement with independent solution experiments for the oligonucleotides containing pyrimidine bases. Although the proposed corrections do not seem to be portable to generic RNA systems, the simulations reveal the importance of the α and ζ backbone angles for the modulation of the RNA conformational ensemble. The correction protocol presented here suggests a systematic procedure for force-field refinement.
