Conclusions

Abstract

The work described here is certainly tributary to some general difficulties of QSAR: if biological affinity depends on a large number of structural features, as it often does, one needs a corresponding number of predictor variables. If the number of experimental points is not several times larger, the reliability of the model Cregressional equation) obtained is low, perhaps even if most of the challenged predictor variables are finally eliminated because of low significance. The use of test series, with correlation of experimental activities vs. calculated ones by means of the regressional equation determined for a different series of compounds, may be a good auxiliary in determining the real predictive value of the model.