Minimizing a suitable free energy expression is arguably the most common approach in (ab initio) protein structure prediction. The achieved accuracy depends crucially on the quality of the free energy expression in use. Here, we present corrections to existing free energy expressions which arise from the thermal motion of the protein. We (i) devise a term accounting for the vibrational entropy of the protein, and (ii) correct existing potentials for the "thermal smoothing effect". (i) Vibrational entropy is almost always neglected in free energy expressions as its consideration is difficult. This practice, however, may lead to incorrect output because distinct conformations of a protein can contain very different amount of vibrational entropy, as we show for the chicken villin headpiece explicitly [1]. For considering vibrational entropy, we suggest a knowledge based approach where typical fluctuation and correlation patterns are extracted from known proteins and then applied to new targets. (ii) At ambient conditions, time-averaged potentials of proteins are considerably smoother when expressed in terms of the average atom coordinates than the Hamiltonian. This effect caused by thermal motion is referred to as the thermal smoothing effect. The strength of the effect varies strongly between atoms. This allows to increase the accuracy of free energy expressions significantly by subdividing atom species regarding their typical fluctuation behavior inside proteins and assigning time-averaged potentials for the new sub-species independently [2]. [1] M. Goethe, I. Fita, and J.M. Rubi, Vibrational Entropy of a Protein: Large Differences between Distinct Conformations, J. Chem. Theory Comput. 11, 351 (2015). [2] M. Goethe, I. Fita, and J.M. Rubi, Thermal Motion in Proteins: Large Effects on the Time-Averaged Interaction Energies, in revision.