Kiralj R., Ferreira M. M. C., “A PRIORI  MOLECULAR  DESCRIPTORS IN QSAR: A CASE OF HIV-1 PROTEASE  INHIBITOR”. Düsseldorf, Germany, 27/08-01/09/2000: 13th European Symposium on Quantitative Structure-Activity Relationships: Rational Approaches to Drug Design, Book of Abstracts, P.58 (2000). Poster P58. Section: New Developments in Chemometrics.

           A PRIORI  MOLECULAR  DESCRIPTORS IN QSAR: A CASE OF HIV-1
                                                 PROTEASE  INHIBITOR

                                      Rudolf Kiralj and Márcia M. C. Ferreira,
    Rudjer Boškovic Institute, Zagreb, Croatia, and Instituto de Química, Unicamp, SP, Brasil

QSAR  methods   today   include  generation  of   many   molecular   descriptors  which   are  then
chosen as the "best"  or  "most significant"  to describe phenomena  of  intermolecular  interactions.
The generation  of  the  descriptors can  hardly avoid  the  useof  quantum  mechanical  and  other
sophisticated calculations,  which in a certain sense becomes a user’s  black box.  Interpretation  of
the nature and meaning of the"best"  descriptors,  as well as of their combinations like in  Principal
Component Analysis,  can be again  a  black box procedure.  Finally one may wonder,  what  really
are the descriptors which fit so nicely to some activity,  if not some pure mathematical concepts  or
physical properties too complicated  to be mentally decomposed into  the basic forces of  nature.
           This work was inspired with  a COMBINE-QSAR study of  HIV-1 protesase  inhibitor   [M.
Pastor, C. Pérez and F. Gago, J. Mol. Graphics Mod., 15 (1997) 363-371.] using 47 descriptors to
reach  r=0.95 on  32  inhibitors. In  the present  work   some  simple  descriptors are  derived  from
empirical  and  2D  structural  formula (called   a priori  descriptors  since  are  known  before  any
sophisticated calculation), in some cases  "calculated"  by  using a pocket calculator.  Through  PLS
method,  aquite reliable model  is obtained  (r=0.90), which   is  far better than  might be  expected
since no 3D optimized structures  and  a  large  amount   of  calculated  molecular  descriptors were
used.
           The presented  a  priori  approach  by simple molecular descriptors  can be used to build  a
prototype  model for a  QSAR problem  as an initial  model which can  be  gradually enriched  with
various calculated descriptors. Also it can help interpreting phenomena in terms  of  basic  chemical
concepts  (molecular  size  and  shape,  chemical  bonds,  atomic  properties,  electron  distribution,
effective  surface  of  ligands  etc.
                                                                                                                                   FAPESP