Modeling purines and pyrimidines with the linear combination of connectivity indices - Molecular connectivity ''LCCI-MC'' method

A series of five experimental properties of DNA-RNA bases (U, T, A, G, and C): singlet excitation energies Delta E(1) and Delta E(2), oscillator strengths f(1) and f(2), and molar absorption coefficient epsilon(260) plus three experimental properties of a wider set of purine and pyrimidine bases: average (pK) = 1/2 (pK(a) + pK(b)), molecular weights MW, and solubility have been simulated in two different ways with linear combinations of connectivity indices (LCCI) chosen from a medium sized molecular connectivity {chi} = {D,D-v,(0) chi,(0) chi(v),(1) chi,(1) chi(v),chi(t),chi(t)(v)} set. A forward selection technique and a full combinatorial space technique have been used to choose the best linear combination of connectivity indices for an optimal modeling. The given properties are very well modeled with the only exception being (pK), whose modeling could be improved with the introduction of fragment reciprocal connectivity indices, that take into account the number of basic and acid groups of the given molecules. The (easier to perform) forward selection technique is in many occasions a good alternative to the more cumbersome full space selection technique and can normally be used to restrict the dimension of the full combinatorial space, thus, facilitating the computation. Limits in the forward selection method can frequently be overcome with the introduction of orthogonal indices. While the simulation of the molecular weights cast some light on the modeling of hydrogen-rich or -poor molecules, the simulation of the solubility shows (i) how far a satisfactory modeling of a small number of compounds can be extrapolated by the aid of the same indices to a wider set, (ii) the importance of linear combinations of squared connectivity indices used in the absolute value mode, and (iii) the contribution of supramolecular connectivity indices for an improved modeling of the solubility. The positive role of the chi(t) and chi(1)(v) indices, all along the modeling of the different properties, seems to be due to the rather low collinearity of these indices relative to the other indices of the connectivity set, a fact that underlines their use in molecular modeling with linear combinations of connectivity indices. In an Appendix, the notation of delta cardinal number is extended to the triplet code words to generate the different families and subfamilies of the genetic code.