Motivation: The creation of artificial proteins is a great challenge in today’s biology. Prediction of the experimental results for changes in proteins surely can considerably accelerate the development of novel proteins. Results: We have derived rules for the prediction of changes in protein thermodynamic stability upon introduction of single substitution in sequence. Using models of neural networks, backward propagation errors, and the modified KRAB method have established the rules. Based on the methods, we developed software allowing us to predict protein free energy upon single substitutions. In this work, we also compare the results. It was demonstrated that the modified KRAB algorithm, when based on the available data, allowed us to predict changes in thermodynamic stability with higher accuracy compared with back propagation networks.