Journal of chemometrics vol:16 issue:2 pages:75-80
A straightforward method to predict melting temperatures of RNA duplexes by artificial neural network is developed. The network consists of three layers in which there are 12 input layer units (10 nearest neighbors and two terminals), two hidden layer units and one output layer unit (melting temperature). It is found that the predicted data for the test set agree with the measured data quite well (AD = 3.7 K, R-2 = 0.9359, SEP = 4.17 for NN1; AD = 2.85 K, R-2 = 0.8843, SEP = 3.43 for NN2). This method omits the determination of thermodynamic parameters and therefore is a simpler way to predict the stability of RNA duplex. Copyright (C) 2002 John Wiley Sons, Ltd.