RNA-binding function prediction of proteins based on their physicochemical features using the logistic regression method

Charge، dipole moment and quadrupole moment eigenvalues as physicochemical features have been used for the RNA-binding function prediction of proteins، in this study. Simple and efficient logistic regression method was utilized for the prediction process. In the corresponding logit equation، charge feature had the highest coefficient (19. 19) and impact on the prediction and dipole moment was the second significant feature. Logistic regression was trained using jackknife procedure on 2601 protein chains (160 RNA-binding proteins and 2441 non RNA-binding proteins). The value for the performance measure of area under the curve of receiver operating characteristics (ROC) was 83% for the final model and is higher than the value obtained by the neural network method for prediction. The values of accuracy، precision and F-measure were 94%، 59% and 46%، respectively، which outperformed the neural network method. In conclusion، we showed that with the help of simple، fast and accurate logistic regression method، RNA-binding proteins can be well distinguished from non RNA-binding proteins using a few number of physicochemical predictor features.