Biomedical Engineering Reference
In-Depth Information
Prediction of Protein Function
HonNianChua
Data Mining Department, Institute for Infocomm Research, Singapore
9.1 Introduction
Automated protein function prediction (PFP) has gained momentum over the last decade
with the proliferation of genomic data. While genomic information becomes available at
a rapidly increasing pace, understanding of the functional mechanism of genes and their
protein products is relatively less efficient. This has motivated computer scientists and biol-
ogists to guide functional discovery by using computational methods to construct functional
models and associations using annotated genes in well-studied model organisms and other
available biological evidence. With limited success, many computational approaches have
been developed to predict gene/protein functions based on a myriad of genomic and experi-
mental evidence, and some of these are readily available for use. In this chapter we provide
a concise overview of these
in silico
PFP methods and elaborate the application of a handful
of these.
9.2 Methods and approaches
Automated PFP has been, and remains, a popular and important area of research in bioin-
formatics and computational biology. A considerable number of approaches have been
explored, which differ in both the use of computational techniques and biological data.
Since a major consideration that guides the choice of an appropriate approach is the avail-
ability of data, we group PFP methods according to the type of biological data used as input
for inference.
As the focus of the chapter is on the practical application of PFP, we limit the scope
to methods that utilize sequence homology [1-3], phylogenetic relationships [4, 5],
sequenced-derived functional and chemical properties [6] and protein-protein interaction
maps [7-15]. These by no means encompass all available PFP approaches, but are
