Assessment of dispersion corrections in DFT calculations on large biological systems

The performance of empirical dispersion corrections in DFT calculations has been assessed for several large, genuine biological systems that include MbAB, H64L(AB), and V68N(AB) (AB?=?CO, O₂), where Mb stands for a wild-type myoglobin, H64L is the (histidine64?→?leucine) mutated myoglobin, and V68N is the (valine68?→?asparagine) mutated myoglobin. The effects of the local protein environment are accounted for by including the five nearest surrounding residues in the calculated systems and they are examined by comparing the binding energies of AB to the myoglobin and to the porphyrin (Por) without residues. Three versions of Grimme's dispersion correction methods, labeled as DFT-D1, DFT-D2, and DFT-D3, were all tested. In the first version (-D1), the dispersion correction (E_disp) is calculated only for noncovalent interactions between molecular fragments and E_disp within a covalent molecule is not calculated. For the DFT functionals, for which the calculated Por–AB binding energies are already too large, only further overestimation occurs when a dispersion correction is made. The geometry optimizations show that the DFT-D2 and DFT-D3 approaches give too short distances between the residues and the heme moiety in the myoglobins and their calculated relative binding energies ΔE_bind(myoglobin-AB/Por–AB) are in poor agreement with experiment in most cases. DFT-D1 performs very well, ensuring structural and energetic features in close agreement with experiment.