During protein folding, protein obtains its characteristic conformation, which is a requirement for function. The rate of folding is typical for each protein. Different variations in proteins can affect the folding rate, changes range more than five orders of magnitude. PON-Fold is the first machine learning method for folding rate predictions for amino acid substitutions. LightGBM algorith was used to develop a gradient boosting method based on known effects on folding rates. The method can be used to predict either named substitutions in one or several proteins or all possible single amino acid substitutions in one protein. A manuscript describing the method is under construction.