Computer-Aided
Formulation – Myth or Reality
Professor Raymond C Rowe
Chief
Scientist, Intelligensys Ltd, Springboard Business
Centre, Ellerbeck Way, Stokesley, North Yorkshire TS9
5JZ, UK.
E-mail:
rowe@intelligensys.co.uk
The development of a
commercial product whether it be a relatively simple
formulation (eg a capsule, tablet or oral liquid) or a
controlled release formulation (eg an implant) is always
a time-consuming and complicated process. Generally an
initial formulation consisting of one or more drugs
mixed with various ingredients (excipients) is prepared,
and, as development progresses, the choice of these and
their levels as well as the manufacturing process are
changed and optimized as a result of intensive,
time-consuming experimentation. These iterations, in
turn, result in the generation of large amounts of data,
the processing and understanding of which is
challenging. In reality, the formulator has to work in a
design space that is multi-dimensional and virtually
impossible to conceptualize.
Traditionally,
formulators have tended to use their own experience in
the generation of the initial formulation and
statistical techniques such as a response surface
methodology to investigate the design space but
optimization by this method can be misleading especially
if the formulation is complex. Nearly twenty years ago
a small number of visionary scientists from both
industry and academia started to address these problems
by experimenting with what was then a relatively well
developed field of computer science - artificial
intelligence. The technologies included expert systems
for the generation
of the initial
formulation; neural networks for modeling the design
space; genetic algorithms for optimizing the formulation
and manufacturing process and neuro-fuzzy logic and
decision trees for exploring the relationships within
the design space and generating understandable rules
that can be used in future work.
Initial results with
expert systems reported in the mid-1990s showed that the
technology did work for tablets, capsules, parenterals,
film-coatings and topicals. The formulations generated
were comparable with those developed by experienced
formulators with the added benefits of consistent
decision making, decreased timelines and cost savings in
drug and excipients1. Slightly later results
with neural networks showed that they could be applied
to any formulation, no matter how complex, provided
there were sufficient data available covering the design
space. They were able to model these data, in many cases
better than statistics, with the added ability of being
either combined with genetic algorithms to accommodate
constraints and preferences in the optimization of a
formulation and process or with fuzzy logic to generate
understandable rules2.
Of course, as with all
new technologies, despite the benefits, there have been
issues, not least related to software and lack of
development skills (the initial problems related to
computing power have been solved with the dramatic
increase in the speed of new computer chips). These have
recently been addressed and there are now neural
computing software packages specifically aimed at
formulators allowing them to easily apply the technology
to problems without having to be an expert in it. These
are increasingly being used in both industry and
academia to model and understand even the most seemingly
intractable problems, e.g., relating the in vivo
response of an inhalation product to its formulation. A
decision support software package for formulators has
also been developed.
The next generation of
formulators in the pharmaceutical industry is likely to
find itself using artificial intelligence technology
routinely and to an increasing extent. Several companies
have already implemented some of the techniques and made
them available either as stand-alone software or linked
via an intranet. However, the largest benefit in the
future will undoubtedly arise from the seamless
integration of artificial intelligence with another
advanced computer technique, that of computer simulation3
into a common decision support system allowing the in
silico generation of formulated products ab
initio.
References
1. Rowe
RC, Roberts RJ. Intelligent Software for Product
Formulation. Taylor and Francis, London, 1998.
2. Colbourn EA, Rowe RC. Neural Computing and Formulation Optimization. In: Swarbrick J, Boylan JC (eds). Encyclopedia of Pharmaceutical Technology, 2nd Ed, Marcel Dekker, 2005.
3. Rowe
RC, Colbourn EA. Computers in Pharmaceutical
Formulation. In: Elkins S (ed). Computer Applications
in Pharmaceutical Research and Development, Wiley-
Interscience, 2006, pp 679-701.
