Communications of the ACM

powerful approach. We are presently
working on building out a ‘reaction re-
pository’ as part of our development of
our RSC data repository and we will be
encouraging the community to contrib-
ute their reaction data.”
The Dial-a-Molecule project is not
the only effort focused on finding ways
to more quickly synthesize molecules
that can be used in research, develop-
ment, and manufacturing processes.
Bartosz Grzybowski, a chemist at
Northwestern University in Evanston,
IL, is working on a synthesis machine
of his own based on Chematica, a soft-
ware/database that uses algorithms
and a collective database of 250 years
of organic chemical information to
predict and provide synthesis pathways
for molecules. Chematica supports 3D
modeling of individual molecules, as
well as labeling of functional groups,
and Grzybowski is negotiating with El-
sevier to incorporate the program into
its Reaxys database, and also is said to
be bidding for a $2.3-million grant from
the Polish government to use Chemati-
ca as the brain of a synthesis machine
that can plan and execute the synthesis
of at least three drug molecules.

Despite the obvious benefits of an
organic synthesis machine, it could be
years before one actually comes to frui-
tion, according to Whitby, who notes
that less than $100,000 of the Dial-A-
Molecule funding grant went to actual
research, with the bulk of the money
used to “identify how we might get to
the target and the key challenges.”
“The Grand Challenge has a 30–
40-year estimated delivery time, so
completion is not imminent,” Whitby
says, contrasting it with large projects
that had a fixed, tangible goal, such
as landing on the Moon. However, he
notes that achievements made over the
next 30 or 40 years on the path to the
development of an organic synthesis
machine likely will have a substantial
impact on chemistry specifically and
our world in general.

Still, while Grzybowski did not respond to a request for comment for this article, he has been quoted as stating that an organic synthesis machine could be built and available within five years. Because he has been shopping Chematica to various entities, few independent assessments of Grzybowski’s efforts have been conducted.

gether so the material can be routed as needed. While Whitby says that from a hardware engineering perspective, such a machine likely could be constructed today (albeit very expensively), it is the software and analytical components of a machine that have yet to be successfully worked out.

The software used in an organic synthesis machine likely would access databases containing information on chemical compounds and their respective properties, as well as the results from chemical reactions that have been conducted and cataloged by the chemistry community. By using these data pieces, the software would be able to accurately combine materials and automatically produce new molecules with a high degree of accuracy and very little human interaction.

The key issue on the software side revolves around figuring out a way to accurately and efficiently apply the various rules and models that govern the way materials interact in combinational chemistry. The sheer number of rules and models can vary widely based on the raw materials used, as well as the specific combinations of these raw materials, thereby adding significant complexity to a potential machine. In essence, the machine would need to calculate the result of each combination of materials, and then ensure the desired rule or model governing the combination was used, which could result in hundreds or thousands of permutations per combination.

Antony J. Williams, vice president of Strategic Development for the Royal Society of Chemistry (RSC) and leader of the Society’s Cheminformatics team (which is working on a collection of reaction data located within its Chem Spider chemical-structure database that will be hosted within the society’s developing data repository), notes that this information is key to the development of a machine capable of fully automating the organic synthesis process.

“I am assuming that the machine
would be underpinned by a strong soft-
ware platform that would utilize some
form of retrosynthetic analysis using
rules extracted from a reaction data-
base,” Williams says. “Basic rules will
certainly get you some way along the
path, but a large database combined
with extracted rules is likely the most
“I have to believe that it is the chem-
istry itself that will be the largest limi-
tation, [with] kinetics of reaction,
side-products and issues such as pre-
cipitation/crystallization,” Williams
says. “I remember trying to do flow-
kinetics in an NMR (nuclear magnetic
resonance) probe, only to have solid
drop out and clog the lines.”
Indeed, work is being done to
smooth this process. Jamison Research
Group, led by Massachusetts Institute
of Technology chemistry professor Tim
Jamison, is working on continuous-
flow synthesis methods, through which
reactions occur as the chemicals move
through a machine (rather than in a
step-by-step process), which can im-
prove speed and yields. This type of
continuous-flow reaction process is
better suited to automation, and could
be integral to the efficient and error-
free design of a fully automated organic
synthesis machine.

Furthermore, Williams notes the overall success of any future organic synthesis machine is predicated on the quality of the underlying reaction databases and the various rules or algorithms used to govern the choice of chemical reactions that can be performed.

“Any predictive algorithm, especially for retrosynthetic analysis, is massively influenced by the underpinning training set and extracted models,” Williams says, which often renders an imperfect end result. Williams says any machine capable of conducting organic synthesis likely will require some form of self-learning capability, so it can grow more efficient over time.