Models for prediction of single base primer extension efficiency from position and type of single mismatch in primer-template duplex

Amplification and specificity of polymerase chain reaction (PCR) are affected by the position and type of primer-template mismatches (MMs) as well as various conditions of reaction. In this study, multiple linear regression (MLR) models and artificial neural network (ANN) models were developed for the prediction of the effects of primer-template mismatches on the primer extension efficiency in primer-template duplex. In MLR models, the independent variable P(i) representing the position effect of i-th mismatch from 3′ end of primers was normalized to values between 0 and 1 according to the size of ΔΔG(i), the difference of Gibbs free energy changes between the mismatch and its corresponding perfect-match, and other independent variables P(j) representing the position effect of the j-th perfect-match from 3’ end of primer were coded 1. A dependent variable of MLR model was relative extension efficiencies of primers. In ANN models, an input layer has neurons equal to the number of independent variables of corresponding MLR models and a hidden layer and an output layer have four and one neurons, respectively. Our MLR models and ANN models outperform the previous polynomial regression model for the prediction of the single base extension (SBE) efficiencies of single-MM primers. Especially, ANN model 6 which has 32 neurons representing the position effect of mismatch, the type of mismatch and the annealing temperature on primer-template duplex in the input layer can predict the SBE efficiencies of single-MM primers with a high accuracy, since its correlation coefficients R in training set, testing set and all data are 0.9870, 0.9782 and 0.9857, respectively. These results will have a good prospect applicable to the design of primer and testing the primer specificity in genome database.