Hansen solubility parameters (HSP) provide essential
information
on the nature of solvents, being a useful tool in their selection
for product and process design. In this work, linear models were developed
to estimate HSP based on the use of COSMO-RS (conductor-like screening
model for realistic solvents) descriptors. Hansen solubility parameters
for 195 compounds were obtained from the literature and classified
into two categories: training set (133 compounds) and testing set
(62 compounds). Then a factorial regression was carried out to predict
dispersion, hydrogen-bonding interaction, and polar contribution of
HSP (δD, δH, and δP, respectively) based on σ-moments and energy COSMO-RS descriptors.
The description of the dispersion contribution was the least successful
due to the low variability of the experimental data, despite the database
containing compounds of a widely different chemical nature. The models
obtained proved to be more than simple mathematical equations since
a physical meaning is achieved when COSMO-RS descriptors are used.
The models developed were then applied to the test database. An excellent
performance was observed, with δH showing the highest R
2 (0.90), and the MAE obtained were 0.98, 1.74,
and 1.44 MPa1/2 for δD, δH, and δP, respectively, which are lower than those
found in previous works. Finally, HSP data for caffeine, nicotine,
paracetamol, and d-camphor were estimated. The results were
close to the literature data and those predictive by the HSPiP software.
Moreover, it was shown that δP and δH drive the molecules’ polarity, impacting their log K
ow (a quantitative measure of polarity). This
work shows that COSMO-RS descriptors are a tool to predict HSP through
linear models, opening new doors in the screening, design, and selection
of solvents to be used in chemical processes.