Previous - Next - Bottom - PP home - PP help TOC

PP Help 08: Methods

Contents

  1. MaxHom : multiple alignment
  2. ProSite : functional motifs
  3. SEG : composition-bias
  4. ProDom : domain assignment
  5. PHDsec : secondary structure
  6. PHDacc : solvent accessibility
  7. Globe : globularity of proteins
  8. PHDhtm : transmembrane helices
  9. Coils : coiled-coil regions
  10. Topits : fold recognition by threading


PREDICTION METHODS

Multiple sequence alignment (MaxHom)

The multiple sequence alignments is built up in essentially three steps (MaxHom, Sander & Schneider, Proteins, 1991, 9, 56-68).
  1. The protein database (currently SWISS-PROT) is searched by a fast alignment program (currently BLASTP).
  2. In sweep 1, sequences are aligned consecutively to the search sequence by a standard dynamic programming method. After each sequence has been added a profile is compiled, and used to align the next sequence.
  3. In sweep 2, after all sequences with significant homology have been picked from the BLASTP output, the profile is recompiled, and the dynamic programming algorithm starts once again to align consecutively the sequences, this time using the conservation profile as derived after completion of sweep 1.


    Functional sequence motifs (ProSite; example for output)

    The following description is from the original ProSite site:
    ProSite is a method of determining what is the function of uncharacterized proteins translated from genomic or cDNA sequences. It consists of a database of biologically significant sites, patterns and profiles that help to reliably identify to which known family of protein (if any) a new sequence belongs.


    Low-complexity regions (SEG; example for output)

    The following description is from the original SEG documentation (JC Wootton & S Federhen, 1996, Meth Enzymology, 266, 554-571):
    SEG divides sequences into contrasting segments of low-complexity and high-complexity. Low-complexity segments defined by the algorithm represent "simple sequences" or "compositionally-biased regions".
    Locally-optimized low-complexity segments are produced at defined levels of stringency, based on formal definitions of local compositional complexity. The segment lengths and the number of segments per sequence are determined automatically by the algorithm.


    Domain assignment (ProDom; example for output)

    The following description is from the original ProDom site (which supplies a rather useful graphical interface to the ProDom database):
    The ProDom protein domain database consists of an automatic compilation of homologous domains detected in the SWISS-PROT database by the DOMAINER algorithm (ELL Sonnhammer & D Kahn, Prot. Sci., 1994, 3, 482-492). It has been devised to assist with the analysis of the domain arrangement of proteins.
    ProDom `domains' are inferred on the basis of conserved subsequences as found in various proteins. Such a conservation corresponds frequently, though not always, to genuine structural domains: therefore domain boundaries should be treated with caution. For some domain families experts have been asked to correct domain boundaries on the basis of both sequence and structural information. This expertise will complement the automated process and improve the quality of ProDom domain families.


    Secondary structure (PHDsec)

    Secondary structure is predicted by a system of neural networks rating at an expected average accuracy > 72% for the three states helix, strand and loop (Rost & Sander, PNAS, 1993 , 90, 7558-7562; Rost & Sander, JMB, 1993 , 232, 584-599; and Rost & Sander, Proteins, 1994 , 19, 55-72; evaluation of accuracy). Evaluated on the same data set, PHDsec is rated at ten percentage points higher three-state accuracy than methods using only single sequence information, and at more than six percentage points higher than, e.g., a method using alignment information based on statistics (Levin, Pascarella, Argos & Garnier, Prot. Engng., 6, 849-54, 1993).
    PHDsec predictions have three main features:
    1. improved accuracy through evolutionary information from multiple sequence alignments
    2. improved beta-strand prediction through a balanced training procedure
    3. more accurate prediction of secondary structure segments by using a multi-level system


    Solvent accessibility (PHDacc)

    Solvent accessibility is predicted by a neural network method rating at a correlation coefficient (correlation between experimentally observed and predicted relative solvent accessibility) of 0.54 cross-validated on a set of 238 globular proteins (Rost & Sander, Proteins, 1994, 20, 216-226; evaluation of accuracy). The output of the neural network codes for 10 states of relative accessibility. Expressed in units of the difference between prediction by homology modelling (best method) and prediction at random (worst method), PHDacc is some 26 percentage points superior to a comparable neural network using three output states (buried, intermediate, exposed) and using no information from multiple alignments.


    Globularity of proteins (GLOBE)

    An additional result from the prediction of solvent accessibility is that of protein globularity. That method is not published, yet. For more information, you may have a look at the preliminary preprint.


    Transmembrane helices (PHDhtm; example for output)

    Transmembrane helices in integral membrane proteins are predicted by a system of neural networks. The shortcoming of the network system is that often too long helices are predicted. These are cut by an empirical filter. The final prediction (Rost et al., Protein Science, 1995, 4, 521-533; evaluation of accuracy) has an expected per-residue accuracy of about 95%. The number of false positives, i.e., transmembrane helices predicted in globular proteins, is about 2% (Rost et al. 1996).
    The neural network prediction of transmembrane helices (PHDhtm) is refined by a dynamic programming-like algorithm. This method resulted in correct predictions of all transmembrane helices for 89% of the 131 proteins used in a cross-validation test; more than 98% of the transmembrane helices were correctly predicted. The output of this method is used to predict topology, i.e., the orientation of the N-term with respect to the membrane. The expected accuracy of the topology prediction is > 86%. Prediction accuracy is higher than average for eukaryotic proteins and lower than average for prokaryotes. PHDtopology is more accurate than all other methods tested on identical data sets (Rost, Casadio & Fariselli, 1996a and 1996b; evaluation of accuracy).


    Coiled-coil regions (COILS; example for output)

    The following description is from the original COILS site:
    COILS is a program that compares a sequence to a database of known parallel two-stranded coiled-coils and derives a similarity score. By comparing this score to the distribution of scores in globular and coiled-coil proteins, the program then calculates the probability that the sequence will adopt a coiled-coil conformation.


    Fold recognition by prediction-based threading (TOPITS; examples for: request and output)

    Remote homologues (0-25% sequence identity) are detected by a novel prediction-based threading method (Rost 1995a and 1995b). The principle idea is to detect similar motifs of secondary structure and accessibility between a sequence of unknown structure and a known fold . For the recognition of similarities between entire folds, the expected accuracy (first hit of alignment list correct) is about 60% (Rost, ISMB95 Proceedings, 1995, AAAI Press, 314-321). If the goal is to correctly detect even short homologous fragments, still about 30% of the first hits are correct (compared to an accuracy of 14% for simple sequence alignments: full paper). Hits with z-scores above 3.0 are more reliable (accuracy > 60%). (Note: a threading service based on similar principles is provided by Daniel Fischer (http://www.mbi.ucla.edu/people/frsvr/frsvr.html).



    Previous - Next - Top - PP home - PP help TOC