Computer- aided formulation development - Computer-Aided Applications in Pharmaceutical Technology - page 23

temperature. Microemulsions and self-microemulsifying drug delivery
systems (SMEDDS) form only in well balanced mixtures of the selected
excipients and within the specifi c concentration ranges of the constituents
at given temperatures and pressures (i.e. the microemulsion area). The
analysis of the infl uence of formulation variables on the area of
microemulsion systems is usually performed within the phase behavior
studies. Pharmaceutically applicable microemulsions consist of fi ve
(surfactant, cosurfactant, oil, water, and drug) or more components.
Complete phase behavior differentiation in such multicomponent
mixtures requires a large number of experiments (Alany et al., 2009;
Friberg and Aikens, 2009). Furthermore, characterization of a
microstructure is a diffi cult task, due to its dynamic character as well as
nanoscale organization (Tondre, 2005).
ANN models were introduced as useful tools for accurate differentiation
and prediction of the microemulsion area from the qualitative and
quantitative composition of different microemulsion-forming systems
(Richardson et al., 1996, 1997; Alany et al., 1999; Agatonovic-Kustrin
and Alany, 2001; Agatonovic-Kustrin et al., 2003; Mendyk and
Jachowicz, 2007; Djekic et al., 2008). The pioneer studies of Richardson
et al. (1996, 1997) demonstrated the use of ANNs to identify the physico-
chemical properties of the cosurfactant with relevance for microemulsion
formation in the four-component system lecithin, (surfactant)/isopropyl
myristate (oil)/triple distilled water/cosurfactant. The different types of
cosurfactants (i.e. short- and medium-chain alcohols, amines, acids, and
ethylene glycol monoalkyl ethers) were employed. The BP feed-forward
algorithm of learning and four computed cosurfactant molecule properties
(molecular volume (v), areas for its head group (a ψ ) and hydrophobe (a φ ),
and computed octanol/water logP value), were selected. The output was
presence (+1) or absence (−1) of microemulsion formation in a particular
mixture.
The data required for ANN training and testing were extracted from
the pseudo-ternary diagrams generated previously by Aboofazeli et al.
(1994), together with the additional data from four pseudo ternary phase
diagrams constructed at a fi xed weight ratio of surfactant-to-cosurfactant
1:1. The trained ANN (the in-house software YANNI) with the
fi nal architecture involving 6 input neurons, a single hidden layer of
14 neurons, and 1 output neuron, was shown to be highly successful in
predicting phase behavior for the investigated systems from the computed
values of v, a ψ , a φ , and logP, achieving mean success rates of 96.7 and
91.6% for training and test data, respectively. These investigations
pointed to the potential of the trained ANN to screen out cosurfactants

￿

￿ ￿