Biomedical Engineering Reference
In-Depth Information
simply needing the structure data, look to other non-experimental methods.
In determining protein structure, the primary alternative to experimental or wet-lab techniques is
bioinformatics. Although computational methods may be able to deliver a solution to a molecular
biology problem such as structure determination in days or weeks instead of months or years, the
solution is only as good as the formulation of the problem. In the case of protein structure
determination or prediction, formulating the problem entails creating a model of the molecule and the
major environmental factors that may influence its structure. With a valid model definition, arriving
at a solution—that is, using the model to drive a simulation of the molecule's behavior and
structure—is simply a matter of executing a program and then evaluating the results.
In order to appreciate the significance of modeling and simulation in bioinformatics, consider that the
first "killer app" on the desktop microcomputer— the one application that raised the status of the
technology from a hobbyist's plaything to a "must have" in business and in the laboratory—was the
now-defunct electronic spreadsheet, VisiCalc. This spreadsheet enabled accountants, engineers, and
physicists to interactively run a variety of what-if scenarios or implicit attempts at problem
formulation to predict the outcomes of virtually any activity that they could express mathematically.
VisiCalc's initial success was due largely to its easy-to-understand user interface of rows and columns
of cells interrelated by position and formulas and a powerful back-end that interpreted the formulas
and graphed the output. Working with Microsoft Excel, Lotus 1-2-3, or other modern electronic
spreadsheets involves creating or using a pre-defined model—a logical, simplified description of how
a real-world system performs. With a valid model—that is, one that adequately describes
relationships in the real world—the spreadsheet provides an environment in which the model can be
brought to life, simulating the activity of the real-world system over time or in response to specific
events. For example, an accountant might look at the expected profits from a business, given a
spreadsheet model that describes sales and business expenses over the course of a year. An
engineer might use a model of a steel beam to explore its dynamic stability when used as a
supporting structure in a bridge. Similarly, a biologist might examine the population dynamics in a
closed ecosystem of various strains of bacteria, based on a model that describes the relationships
between population, food supply, and the environment. A spreadsheet model is a set of linear
equations relating the values of several variables (cells).
Equipped with a spreadsheet and a few equations, a molecular biologist might define a model of a
neural network that can learn to recognize amino acid sequences and assign protein structures to
certain sequences, as in
Figure 9-1
. The model of a single neuron in the artificial neural network
defines the output of the neuron as the weighted sum of inputs to the neuron, including feedback
from the output: The model of the entire neural network additionally specifies the interconnection of
the individual neuron models. Mathematically, the model of an individual neuron that can accept four
outputs (
o
) with their associated fixed weights (
w
) can be expressed as:
Figure 9-1. Model of a Single Neuron. This model is used in the simulation
of a neural network (inset) that can be used to classify patterns, such as
protein structures associated with specific amino acid sequences. The
model and associated simulation can be created in a general-purpose
spreadsheet or in a computational environment specifically designed for the
simulation of neural networks.
