Raport de cercetare - Lorentz JÄNTSCHI

allowed us to achieve a precise and relatively fast method for determination of log Po/w of
different series of drug compounds and to predict with suf.cient accuracy the log Po/w of new drug
derivatives.
÷ Lucrare: A multiple linear regression and partial least squares study of flavonoid compounds
with anti-HIV activity
÷ Autori: C. N. Alves, J. C. Pinheiro, A. J. Camargo, M. M. C. Ferreira, R. A. F. Romero, A. B. F.
da Silva
÷ Sursa: Journal of Molecular Structure: THEOCHEM, 2001,Volum 541(1-3), p. 81-88
÷ Rezumat:
The molecular orbital semi-empirical method PM3 was employed to calculate a set of molecular
properties (variables or descriptors) of 21 flavonoid compounds with anti-HIV activity. The
correlation between biological activity and structural properties was obtained by using the
multiple linear regression and partial least squares methods. The model obtained showed not only
statistical significance but also predictive ability. The significant molecular descriptors related to
the compounds with anti-HIV activity were: electronegativity (χ) and the charges on atoms C3
and C7 (Q3 and Q7, respectively). These variables led to a physical explanation of electronic
molecular property contributions to HIV inhibitory potency.
÷ Detalii de interes:
Significant regression equations were obtained by multiple linear regression and partial least
squares methods for 20 flavonoid compounds according to their anti-HIV activity. The best
regression equation obtained was based on the following descriptors: electronegativity (χ) and
atomic charges on atoms C3 and C7 (Q3 and Q7, respectively). The model obtained showed not
only statistical significance but also predictive ability and revealed that higher values for χ
combined with high positive charges on C7 and high negative charges on C3 lead to an
increasing of the anti-HIV activity. These variables allowed a physical explanation of electronic
molecular properties contributing to HIV inhibitory potency as the electronic character relates
directly to the electron distribution of interacting molecules at the active site.
A comparison of the performance between the MLR and PLS models showed the PLS have
substantially better predictive capability than the MLR model, even thought their correlation
coefficients are comparable. This indicates clearly that the correlation coefficient by itself is not a
good parameter for testing the model performance. Also, it has been shown that PLS is an
excellent tool for those cases where the descriptors are by any means correlated.
÷ Lucrare: A faster algorithm for ridge regression of reduced rank data
÷ Autori: Douglas M. Hawkins, Xiangrong Yin
÷ Sursa: Computational Statistics & Data Analysis 40 (2002), p. 253 - 262
÷ Rezumat:
Regression data sets typically have many more cases than variables, but this is not always the
case. Some current problems in chemometrics-for example fitting quantitative structure activity
relationships-may involve fitting linear models to data sets in which the number of predictors far
exceeds the number of cases. Ridge regression is an approach that has some theoretical foundation
and has performed well in comparison with alternatives such as PLS and subset regression. Direct
implementation of the regression formulation leads to a O(np 2 +p 3 ) calculation, which is
substantial if p is large. We show that ridge regression may be performed in a O(np 2 )
computation-a potentially large saving when p is larger than n. The algorithm lends itself to the
use of case weights, to robust bounded influence fitting, and cross-validation. The method is
illustrated with a chemometric data set with 255 predictors, but only 18 cases, a ratio not unusual
in QSAR problems.
÷ Detalii de interes:
The residuals from the ridge regression have variance (1−hi)σ2, and so internally studentized
residuals can be computed in the usual way from the ridge residuals, leverages, and setting up the
ridge regression with weights leads to a number of possibilities.
34