Using Artificial Intelligence in Chemistry and Biology: A Practical Guide
CRC Press, a division of
Taylor & Francis in May, 2008.
1. An Introduction to Artificial Intelligence
2. Artificial Neural Networks
3. Self-Organizing Maps
4. Growing Cell Structures
5. Cellular Automata
6. Evolutionary Algorithms
7. Expert Systems
Particle Swarm Optimization
8. Fuzzy Logic
9. Classifier Systems
10. Evolvable Developmental Systems
In each area, the principles of the method are discussed, together with an outline of how
the method can be used to solve scientific problems. The text includes a CD.
Introduction to Artificial Intelligence
It is clear that many problems in science are challenging to solve. A typical example is the task
of finding a new drug to treat a defined disease.
The number of compounds that could conceivably act as drugs is not infinitely large (many
drugs act by locking into the active site of a protein, so need to be quite small,
otherwise they would not fit into the active site).
Nevertheless, the number of possibilities is substantial, roughly 1040.
To assess such a large number of possibilities computationally, smart
methods are needed to sift out potentially valuable drugs from all the rest.
Artificial Intelligence offers tools that can help in this task.
Artificial Neural Networks
Artificial Neural Networks (ANNs) are probably the most widely used AI tool in
the experimental sciences, since they can be trained to accomplish a considerable range of tasks.
An ANN consists of a number of simple computational units which, on their own, can
accomplish rather little, but may have great power when joined in a network.
The role of a self-organizing map is to crush multi-dimensional data down into a smaller
number of dimensions (usually two), so that the clustering of datapoints in n-dimensional space
is made apparent.
Here are two images drawn from the development of a Self-organizing map:
The figure on the right shows the structure of a self-organizing map mid-way through training.
The colours are determined by interpreting the first three weights at each node as RGB
values. On the left, the same set of weights is plotted, but this time the colours are determined by
how different the weights at one node are from the weights in the small group of
nodes that immediately surround it.
Self-organizing maps are widely used in science to group and classify data, particularly when the
data are extremely complex, such as the mass spectra of multi-component mixtures
like oils or biological samples.
Cellular Automata (CA) models are straightforward to use and intuitively simple. They break a region of simulation
into a number of discrete regions, replacing the differential equations which would often be
used to model dynamic systems by a large number of small, simple units. The most widely-known
example of a CA model is probably John Conway's Game of Life.
Cellular Automata show potential in science in the modelling of systems that
change in space, time, or both, such a bacterial colonies (below).
EJS - Easy Java Simulations
EJS ("Easy Java Simulations") has been used to prepare a number of the illustrations in the book.
This tool, written by Prof Francisco Esquembre at the University of Murcia in Spain, provides a
simple but powerful way to introduce professional graphics into Java. A full, free, version of EJS
is provided on the CD that accompanies the text.
A screen shot from a typical EJS application - a simulation of particles in a three-dimensional
box showing velocity traces and calculated forcefield - appears below.
Updated November 19, 2008.