Melting Point and Fusion Enthalpy QSPR Models

In summary, this workflow constitutes an efficient and probably one of the most accurate ways to predict solubilities. Moreover, it is applicable also for nonexperts in the field of computational chemistry. 

5 MELTING POINT AND FUSION ENTHALPY QSPR MODELS 

In order to develop a melting point model a dataset was created from prior published sources. Melting points from the PHYSPROP database [40] and from the work of Karthikeyan and coworkers [41] were combined. However, we had to find out that this combined dataset contains a significant amount of noise, plainly speaking basically questionable entries. Thns, suspect data points were ranoved or corrected resulting in a dataset containing the melting point and SMILES of more than 12,(XX) compounds. 

As a test set 277 compounds were kept aside, corresponding to the data used by Karthikeyan and coworkers as a test set. 

For this specific melting point model 14 variables have been selected by iterative forward selection (Fig. 9.4), yielding an overall accuracy (RMSE) after fivefold cross-validation of 41.2 K (Table 9.4). 

---