Summary The paper compares linear, quadratic, and cubical regression together with several weighted and robust approaches in the context of lipophilicity determination. The comparison is done on 35 model compounds on data from different modifiers used on RP18, CN, and silica plates. It can be concluded that the use of weighted and moderately robust regression technique increases correlation between extrapolated retention and real lipophilicity, whereas polynomial and very robust techniques give visibly worse results due to their excessive flexibility and higher extrapolation uncertainty. Additionally, we have compared averaging retention from different modifiers by RF, k, and RM values. The results are similar; however, surprisingly, RF averaging performs slightly better to the other approaches.
Read full abstract