QSAR modeling of dihydrofolate reductase inhibitory activity by correlation weighting of nearest neighboring codes

The nearest neighboring code (NNC) is a local graph invariant. The NNC of a given vertex of the labeled hydrogen filled graph (LHFG) is a function of atom composition of the vertex neighbors. By optimization correlation weights of local invariants of the LHFG one-variable models of dihydrofolate reductase inhibitory activity have been obtained. As local graph invariants Morgan extended connectivity of zero, first, and second orders as well as the NNCs have been used. Best such model has been obtained with the NNC. Statistical characteristics of the model are the follows: Training set: n=33, r2=0.9273, s=0.155, F=395; Test set: n=35, r2=0.8782, s=0.259, F=238; All structures: n=68, r2=0.8855, s=0.213, F=511. [Copyright &y& Elsevier]