Reasonable QSAR model, i.e. constructing quantitative structure-activity relationship on the basis of known ligand/receptor interaction mode, was proposed in this paper, which can avoid defections of traditional methods that build prediction model depending upon only structural features of molecules themselves. When applying it to study on peptide/protein affinity activities, a new method of genetic virtual screening of combinative mode for peptide/protein (GVSC) was proposed with genetic algorithm as screening tools for combinative targets and interaction mode of virtual receptors. GVSC technique successfully solved problems in reasonable QSAR study that it is hard to know ligand/receptor interaction mode under conditions of unknown structures. Further, 58 angiotension converting enzyme and 18 double site mutation residues in camel antibody protein cAb-Lys3 were employed respectively to examine the GVSC model. It was suggested that the GVSC model should outperform traditional QSAR model to interpret ligand/receptor interaction mode.

Key words: genetic virtual screening of combinative mode of peptide/protein, quantitative structure-activity relationship, genetic algorithm, partial least square regression