Biomedical Engineering Reference
In-Depth Information
True ab initio methods of predicting the structure of large protein molecules, which are based on
modeling of atomic-level forces, are beyond the capabilities of even the fastest supercomputers. For
all but the smallest protein molecules, true ab initio modeling will have to wait for affordable, higher-
performance computing.
Issues
Two major concerns in modeling and simulation in bioinformatics are consistency and performance.
Consistency is an issue in tasks such as protein structure prediction because different assumptions
and simplifications result in different protein structures, even with the same source data. For
example, if two different computer systems use the Metropolis Algorithm to solve for the global
minimum of a function used to define protein structure, and one system provides a more random
distribution without major holes in the distribution while the other system produces much more
pseudorandom data, each system will produce different analytical results. As a result, two different
tools designed to predict protein structure will likely predict two different structures, for example. The
difference in the predicted structures may be significant, depending on how poorly the pseudorandom
number generator on the inferior system performs, and how reliant the system is on the Metropolis
Algorithm for predicting protein structure. As such, it's important for the user to determine the
underlying assumptions and intended purpose of a simulation-based tool before blindly relying on it
to provide credible results.
As noted in the discussion of hardware, performance is always an issue in modeling and simulation
systems intended to work with large data sets exemplified by protein data. Many commercial
programs claim significant performance advantages over the academic versions of the same program
due to more efficient coding or the use of proprietary algorithms. Whether the expense of commercial
systems is warranted depends the amount of modeling and simulation work routinely performed in
the course of R&D, the available hardware, and cost considerations.
However, even with an unlimited hardware budget, it's currently impossible to simulate the
interaction of each of the thousands of atoms in a large protein molecule with each of the hundreds
of thousands of atoms in the surrounding aqueous environment. Assumptions and simplifications
have to be made to realize reasonable performance, and these simplifications and assumptions will
inevitably have an adverse affect on consistency.
