科研之友
微信
新浪微博
Facebook
分享链接
摘要
A new descriptor, namely scores vector of zero dimension, one dimension, two dimension and three dimension (SZOTT), was derived from principle components analysis of a matrix of 1 369 structural variables including 0D, 1D, 2D and 3D information for 20 coded amino acids. SZOTT scales were then employed to express structures of 20 thromboplastin inhibitors and 34 bactericidal peptides. The correlation coefficients of both whole calibration (R-2 = R-cu(2)) and of cross validation (Q(2) = R-cv(2)) for the multiple-variable models by classical partial least squares (PLS) and orthogonal signal correction-partial least squares (OSC-PLS) of 20 thromboplastin inhibitors were 0.989 and 0.748, 0.994 and 0.936, respectively. R-2 and Q(2) for the models by PLS and OSC-PLS of 34 bactericidal peptides were 0.619 and 0.406, 0.910 and 0.503, respectively. Satisfactory results obtained showed that structural information related to biological activity in both data sets could be described by SZOTT which included plentiful information related to biological activity, and which was conveniently operated and easy interpreted., also predictive capability of models were relative robust. There is a high prospect for SZOTT wide applications on quantitative sequence-activity modeling (QSAM) of peptides.
