Quantitative Structure-Property Relationship for Predicting Surface Tension of Organic Compounds Using Associative Neural Networks

This paper explains associative neural network based quantitative-structure property relationship study for prediction of surface tension of organic compounds using molecular descriptors derived-from molecular Structures. A set of 116 organic compounds, which includes 48 alkanes, 31 alcohols, 20 amines, 14 alkenes and 3 aldehydes as data series are selected for the present study. Unsupervised forward selection strategy is used for descriptor selection. from the large, Set of descriptors using E-DRAGON software and six descriptors are selected for model development for surface tension. Associative neural network method is used to construct the non-linear prediction model for Surface tension. The selected descriptors are used as input data for training and testing the associative neural network. The predicted results:are in good agreement with the experimental surface tension, of organic compounds with squared-correlation co-efficient (R-2) of 0.98 for training and 0.932 for testing. The results are cross-validated by leave-one-out procedure. The model is Suitable to a large variety of compounds, which predicts better than other models reported in previous studies.