To date there is only one three-dimensional crystal structure of a GPCR, that of bovine rhodopsin at 2.8A (Palczewski et al. 2000). Although this may be used as a yardstick for assessing the reliability of transmembrane domain (TMD) prediction methods for type I receptors, predictions will have to be relied upon for the foreseeable future. The are many available methods for defining TMD regions based on the primary amino acid sequence (see the ExPASy web site for a list: http:/www/expasy.ch/tools). It is also instructive to look at the predictions in tandem with simple hydrophobicity plots (e.g. pepplot, in GCG) (Kyte and Doolittle 1982) as part of the prediction process.
Good results have been obtained when multiple aligned sequences can be analysed; however, in many cases, predictions on single-sequences is important. The Dense Alignment Surface (DAS) method (http://www.sbc.su.se/~miklos/DAS/) was introduced in an attempt to improve sequence alignments in the G protein coupled receptor family of transmembrane proteins (Cserzo et al. 1994) and as now been generalized this method to predict transmembrane segments in any integral membrane protein. DAS is based on low-stringency dot-plots of the query sequence against a collection of non-homologous membrane proteins using a special scoring matrix. TMPRED (http://www.ch.embnet.org/software/TMPRED_form.html), uses an algorithm based on the statistical analysis of data in TMbase, which are data largely based on SwissProt data on transmembrane proteins. Other tools use machine learning methods, such as TMHMM (http://www.cds.dtu.dk/services/TMHMM/), HMMTOP (http://www.enzim.hu/hmmtop/index.html). These are based on training sets of transmembrane proteins and negative sets of non-membrane proteins. With all methods, improvement should be seen when actual crystal structures of membrane proteins become more common. A modified version of TMHMM for use with GPCRs, 7TMHMM, will soon be published.
In terms of which program performs the best, it is important to remember that all methods are predictions. Furthermore, a high degree of the underlying data are common to all and comes from the expertly analysed sources. All programs have user-friendly web interfaces but it is the display of results that becomes the factor in deciding user preference. Most produce graphic output but SOSUI (http://sosui.proteome.bio.tuat.ac.jp/sosuiframe0.html) (Hirokawa et al. 1998) produces the most extensive (table, hydropathy plot, helical wheel, and a cartoon).
Was this article helpful?