Vol.:(0123456789)
1 3
Journal of Computer-Aided Molecular Design
https://doi.org/10.1007/s10822-018-0125-y
Prediction of partition and distribution coefficients in various solvent
pairs with COSMO-RS
Soa Tshepelevitsh
1
· Kertu Hernits
1
· Ivo Leito
1
Received: 15 March 2018 / Accepted: 25 May 2018
© Springer International Publishing AG, part of Springer Nature 2018
Abstract
Performance of COSMO-RS method as a tool for partition and distribution modeling in 20 solvent pairs—composed of neu-
tral or acidic aqueous solution and organic solvents of different polarity, ranging from alcohols to toluene and hexane—was
evaluated. Experimental partition/distribution data of lignin-related and drug-like compounds (neutral, acidic, moderately
basic) were used as reference. Several aspects of partition modeling were addressed: accounting for mutual saturation of
aqueous and organic phases, variability of systematic prediction errors across solvent pairs, taking solute ionization into
account. COSMO-RS was found to predict extraction outcome for both ligneous and drug-like compounds in various solvent
pairs fairly well without any additional empirical input. The solvent-specific systematic errors were found to be moderate,
despite being statistically significant, and related to the solvent hydrophobicity. Accounting for mutual solubilities of the
two liquids was proven crucial in cases where water was considerably soluble in the organic solvent. The root mean square
error of a priori logP prediction varied, depending mainly on the solvent pair, from 0.2 to 0.7, overall value being 0.6 log
units. The accuracy was higher in case of hydrophilic than hydrophobic solvents. The logD predictions were less accurate,
due to pK
a
prediction being an additional source of error, and also because of the complexity of modeling the behaviour of
ionic species in the two-phase system. A simple correction for partitioning of free ions was found to notably improve logD
prediction accuracy in case of the most hydrophilic organic phase (butanol/water).
Keywords COSMO-RS · Distribution coefficient · Molecular modeling · Liquid–liquid extraction · Solvent effects
Introduction
Partition properties of compounds play important roles
in various branches of research and technology. Partition
coefficients in octanol/water system (logP
o/w
) are the most
widely studied due to their correlation with physicochemical
properties [1–4] and ecotoxicological effects [5–7] of the
compounds. LogP
o/w
is one of the parameters in Lipinski’s
“rule of five” [8], and predicting it with reasonable degree
of accuracy is essential for rational drug design and drug
candidate screening. Modeling the distribution of various
compounds in numerous other solvent pairs is important for
design and optimization of industrial separation processes.
Given the sufficient reliability of the predictions and afford-
able computational cost, computational solvent screening
is a preferable option compared to all-experimental studies.
The field of computational description of liquids is ruled
by semi-empirical methods, since ab initio modeling of a
liquid mixture would be complicated and extremely com-
putationally expensive. Various methods for logP prediction
vary in complexity and accuracy [9, 10]. However, most of
them require parametrization with the experimental data of
the solvent pair of interest, so limited availability of such
data is a serious hindrance to their use.
Conductor-Like Screening Model for Real Solvents
(COSMO-RS) [11–15] is a method combining unimolecu-
lar quantum mechanical calculations and statistical thermo-
dynamics of molecular interactions and is used mainly for
predicting thermodynamic properties of liquid mixtures.
COSMO-RS has several unchallenged advantages in the
field of extraction modeling. First, it allows predicting the
Electronic supplementary material The online version of this
article (https ://doi.org/10.1007/s1082 2-018-0125-y) contains
supplementary material, which is available to authorized users.
* Ivo Leito
ivo.leito@ut.ee
1
Institute of Chemistry, University of Tartu, Ravila 14a,
Tartu 50411, Estonia
@Phil_Robichaud
@deepthiw
@JoseServera