PREDICTING INHIBITION CONSTANTS OF ANTIMALARIAL DRUG COMPOUNDS USING SUPPORT VECTOR REGRESSION

Malaria is one of the deadly infectious diseases in tropical regions. World Health Organization (WHO) reported that this disease caused 3.3 million deaths during 2001-2012. Lethality of Malaria is due in part to the high mutation rate of Plasmodium falciparum,, a common parasite of mosquitos that causes the disease. Consequently, the effectiveness of currently available antimalarial drugs is decreased rapidly. New potent drugs are urgently needed to battle this fatal disease. However, drugs development processes take enormous amount of money and time. In this work, we employ an optimization technique, the support vector regression, to create a prediction model for the inhibition constant, K-i, of Trimethoprim, Pyrimethamine, and Cycloguanil, which are potential antimalarial drug compounds, based on their 248 molecular descriptors. Since the high number of descriptors can be extracted from a drug compound, we employ a feature selection technique based on F score before constructing the prediction model. The results show that a prediction model based on SVR with linear kernel provide the highest accuracy and also outperform existing techniques. Our proposed model yields a strong coefficient of determination R-2 = 0.82 of predicted K-i versus experimental K-i using 5-fold cross validation. The prediction model suggest that K-i is an output of a linear function using only 18 descriptors. Moreover, dipole moment values have highest weights in the final prediction model. This also suggests that K-i values have a positive correlation with dipole moment values. The results of this study demonstrate that low dipole moment values lead to low K-i values. Our proposed prediction model can be used as a screen to eliminate potential drugs with high K-i values for further developments which, in return, may help reduce cost and time for antimalarial drug developments.