USE OF THE MULTIFUNCTIONAL AUTOCORRELATION METHOD TO ESTIMATE MOLAR VOLUMES OF ALKANES AND OXYGENATED COMPOUNDS - COMPARISON BETWEEN COMPONENTS OF AUTOCORRELATION VECTORS AND TOPOLOGICAL INDEXES

The Multifunctional Autocorrelation Method (MAM) derived from the auto correlation method of Moreau et al. was used to calculate molar volume s for 300 linear or branched alkanes from C-6 to C-11 and 104 oxygenat ed compounds including 40 alcohols, 30 ethers and 34 ketones. Componen ts C-i of a MAM connectivity vector were compared with several topolog ical descriptors. MAM descriptors were generally more efficient than t opological descriptors but good regressions were obtained for alkanes only when more than two components C-i were used. An excellent regress ion was obtained with four components (C-0 to C-4): V-m= 4.45C(0) + 6. 44C(1) - 3.33C(2) + 0.46C(4) + 52.9 (n = 300, r= 0.999, s=0.81) Models obtained using descriptors based on connectivity were good enough for many practical purposes. However, to derive models adapted to estimat e molar volumes for alkanes, an autocorrelation vector V based on Van der Waals volumes (as given by Bondi) was considered. The best equatio n obtained again includes three components: V-m = 1.750 V-0 - 0.123 V- 3 - 0.037 V-4 + 17.97 (n = 300, r = 0.999, s = 0.85) To extend the use of MAM to compounds containing heteroatoms, a set of alcohols, ketone s and ethers, where each functional group represents a different inter action type, was considered, Chemical families were first studied indi vidually, then the total set was considered. The components of an auto correlation vector V based on Van der Waals volumes were used as molec ular descriptors. Oxygenated compounds are correctly described by only one component V-0. The correlation coefficient is 0.993 for all the c hemical families studied, and the standard error of the estimate is lo w (s=2.56). Comparison of standard errors of the estimate for Wiener's indices (s=17.38) and autocorrelation components (s = 3.95), for all chemical families, shows that components of MAM are more efficient as molecular descriptors. The quality of regression equations was not sig nificantly changed when 104 oxygenated compounds and 141 alkanes were studied in a global set.