The molecular orbital method PM3 is employed to calculate a set of molecula
r descriptors (variables) for 36 deoxy analogs of 1-[(2-hydroxyethoxy)methy
l]-6-(phenylthio)thymine (HEPT) with anti-HIV-l activity. Pattern recogniti
on methods, principal component analysis (PCA) and stepwise discriminant an
alysis (SDA) were employed in order to reduce dimensionality and investigat
e which subset of variables should be more effective for classifying the HE
PT-analog compounds according to their degree of anti-HIV-l activity. The P
CA showed that the variables log P (partition coefficient), MR (molecular r
efractivity), Delta H-f (heat of formation), Q(N) (net atomic charge on ato
ms 2 and 3), and chi (Mulliken's electronegativity) are responsible for the
separation between compounds with higher and lower anti-HIV-1 activity. By
using the SDA we have found the following descriptors as responsible for t
he separation between the active and less active compounds: log P (partitio
n coefficient), chi (Mulliken's electronegativity), mu (dipole moment), Q(4
) (net atomic charge on atom 4), and t(2) (torsional angle). From the SDA w
e present a prediction rule for classifying new HEPT-analog compounds with
anti-HIV-l activity. (C) 2000 Elsevier Science B.V. All rights reserved.