three 0.749 0.827 0.502 0.678 0.845 0.535 0.685 0.878 0.702 0.828 0.606 0.714 0.452 0.417 0.842 0.608 0.728 221 212 152 220 246 124 160 130 127 117 156 250 146 148 201 143 135 135 155 164 131 573 407 149 183that applying two further charge descriptors from the dissociated molecule can

three 0.749 0.827 0.502 0.678 0.845 0.535 0.685 0.878 0.702 0.828 0.606 0.714 0.452 0.417 0.842 0.608 0.728 221 212 152 220 246 124 160 130 127 117 156 250 146 148 201 143 135 135 155 164 131 573 407 149 183that using two further charge descriptors in the dissociated molecule can markedly boost the predictive power on the EEM QSPR models. Tables 2 and three, Figure 1 show that these new 5d EEM QSPR models deliver much better pKa prediction than their corresponding 3d EEM QSPR models. Specifically, adding the descriptors derived in the dissociated molecules increased the average R2 value for the EEM QSPR models from 0.876 to 0.913.Comparison of EEM QSPR models and QM QSPR modelshave average R2 = 0.951. We also note that adding extra descriptors to a QM QSPR model brings much less improvement than adding much more descriptors to an EEM QSPR model.Influence of theory level and basis setAnother essential question is how correct the EEM QSPR models are in comparison with QM QSPR models. Table 2 and Figure 1 show that QM QSPR models deliver, in most cases, a lot more precise predictions. This really is confirmed also by the typical R2 values from Table three. This is not surprising, given that EEM is an empirical approach which just mimics the QM strategy for which it was parameterized. An intriguing fact is that the differences in accuracy involving QM QSPR models and EEM QSPR models usually are not substantial.1228595-79-6 In stock As an example, 5d EEM QSPR models have typical R2 = 0.913, even though 5d QM QSPR modelsEEM parameters are readily available only for a reasonably modest quantity of theory levels (HF and B3LYP) and basis sets (STO3G and 61G). Therefore we can not perform such a deep evaluation of theory level and basis set influence on pKa prediction capability from EEM atomic charges, as was carried out for QM QSPR models by Gross et al. [22] or Svobodova et al. [24]. We are able to only evaluate the models employing HF/STO3G and B3LYP/61G charges, as these are the only combinations for which EEM parameters are obtainable for the exact same population evaluation (MPA). Consequently we can study only the influence of your mixture of theory level / basis set, and not the isolated influence of your theory level or basis set. Our analysis revealed that B3LYP/61G charges supply slightly extra precise QM QSPR models than HF/STO3G charges (seeSvobodovVaekovet al. Journal of Cheminformatics 2013, 5:18 a r a http://www.jcheminf.com/content/5/Page eight ofQM theory level basis set HF/STO3GPAEEM parameter set nameR two of QSPR model 3d EEM 3d EEM WO 5d EEM 3d QM 5d QM 0.1196145-01-3 Order 8671 0.PMID:24275718 8663 0.8737 0.8671 0.9099 0.8860 0.8696 0.8910 0.8876 0.8731 0.8727 0.8848 0.9044 0.8415 0.8696 0.8639 0.8695 0.8646 0.9239 0.9239 0.9127 0.9241 0.9166 0.9151 0.9182 0.9198 0.9151 0.9043 0.9113 0.9012 0.9098 0.8838 0.9224 0.9053 0.8863 0.8972 0.9179 0.9189 0.9203 0.9179 0.9195 0.9142 0.9154 0.9192 0.9158 0.9094 0.9132 0.8866 0.9180 0.9050 0.9148 0.9131 0.9057 0.9017 0.9515 0.HF/631G B3LY P/631GMPA Svob2007 cbeg2 Svob2007 cmet2 Svob2007 chal2 Svob2007 hm2 Baek1991 Mort1986 Jir2008 hf MK Chaves 2006 Bult2002 mul NPA Ouy2009 Ouy2009 elem Ouy2009 elemF Bult2002 npa Hir. Bult2002 hir MK Jir2008 mk Bult2002 mk Chel. Bult2002 che AI M Bult2004 aim MPA0.8405 0.8865 0.9671 0.9724 0.9590 0.0.9042 0.9477 0.8447 0.8960 0.8528 0.9087 0.9609 0.Legend fantastic pretty great excellent satisfactory acceptable weak R2 0.95 0.97 0.92 0.95 0.91 0.92 0.9 0.91 0.85 0.9 0.8 0.Figure 1 R2 for the correlation in between calculated and experimental pKa .Table 3 Typical R2 amongst experimental and predicted pKa for all.