Predicting the Mechanism of Action for Narcotic and Reactive Compounds

Predicting the Mechanism of Action for Narcotic and Reactive Compounds 

The second experiment we present in this tutorial for classifying compounds according to their mechanism of action involves the classifications of 88 chemicals. The chemicals are either narcotics (nonpolar and polar narcotics) and reactive compounds (respiratory uncouplers, soft electrophiles, and proelectrophiles).The dataset, consisting of 48 narcotic compounds 

The best prediction statistics for each kernel type are presented here linear, C = 1000, ACp = 0.86 polynomial, degree 2, C = 10, ACp = 0.92 RBF, C = 100, y = 0.5, ACp = 0.83 neural, C = 10, a = 0.5, b = 0, ACp = 0.78 and anova, C = 10, y = 0.5, d = 1, ACp = 0.87. These results indicate that a degree 2 polynomial is a good separation hyperplane between narcotic and reactive compounds. The neural and RBF kernels have worse predictions than does the linear SVM model, whereas the anova kernel has similar ACp with the linear model. 

---