QSAR analysis of estrogen receptor ligands using neural networks

Quantitative structure-activity relationship analysis of estrogen receptor ligands were constructed by means of neural networks. Three layer networks were trained with the back-propagation algorithm to predict the relative bi nding affinity of 4-substituted deoxyhexestrol derivatives with estrogen re ceptor in lamb uterine. The relationships between structure and relative bi nding affinity were examined quantitatively using four molecular parameters (ClogP, MR, sigma and sigma (-)). The results obtained were compared to th ose given in the literature by the multiple linear regression, and were fou nd to be better. The contribution of each descriptor to the structure-activ ity relationship was evaluated. Hydrophobicity of the molecule was thus fou nd to take the most relevant part in the molecular description.