On the basis of the values contained in the literature published in 19002016, we have developed an experimental database including sublimation Gibbs energies, enthalpies, and melting temperatures of 1515 compounds. We have also suggested an algorithm of database fragmentation which includes groups/clusters with structurally similar compounds. For this aim we used Tanimoto similarity coefficients. Clusterization was carried out for each substance of the test set. All the points within a cluster were smoothed by a linear function in the coordinates of Gibbs energy vs melting temperature. Using the training and test sets, it has been shown that the algorithm suggested by us describes experimental data well (rms = 3.89 kJ.mol(-1)). We have developed quantitative structure-property relationship models based on HYBOT physicochemical descriptors and melting points in order to predict sublimation Gibbs energies and enthalpies of molecular crystals. The developed approach was applied to determine cocrystal formation thermodynamics. Twenty nine out of 30 seven cocrystals selected in the literature have been predicted correctly (78% of the correct matches).
