[Rejection of physico-chemical indices for pathological matrix in a quantitative drug design].

ABSTRACT

In a quantitative drug design, correlation of coefficient matrix of physico-chemical indices Xi is often pathological when these indices are highly correlated. The regression equation of biological activity Y about these indices Xi obtained in this case is not stable. In this paper, a method is given to obtain a stable regression equation giving a critical valne alpha and finding out the indices among which correlation coefficient is not less than alpha. The following are the rules to reject some of them. If Xi and Xj is highly correlated (magnitude of rij greater than or equal to alpha, rij is the correlation coefficient of Xi and Xj), and magnitude of rin greater than magnitude of rjn (rin and rjn are correlation coefficients of Xi and Xj about Y, respectively), than the index Xj is rejected, otherwise, Xi is rejected. Stable equation can be obtained by stepwise regression with the remaining indices.