Comparison of spectrum-like representation of 3D chemical structure with other representations when used for modelling biological activity

Various possibilities of structure representations are discussed and compared with the new uniform structure code-named spectrum-like representation. The advantages and pitfalls of this new structural representation are deduced from modelling biological activity of a group of 28 flavonoid analogues represented by five different structural codes. The counterpropagation artificial neural network and multiple linear regression are applied as modelling techniques. From the modelling results, it can be concluded that the suggested representation is well suited for QSAR studies. However, additional to pure geometry, electronic structure of the compounds must be included into the representation of molecular structures because it plays an important role in the reaction mechanism inhibition of the enzyme p56lck by binding on the enzyme receptor active site.