BACKGROUND: Knottins are small, diverse and stable proteins with important drug design potential. They can be classified in 30 families which cover a wide range of sequences (1621 sequenced), three-dimensional structures (155 solved) and functions (> 10). Inter knottin similarity lies mainly between 15% and 40% sequence identity and 1.5 to 4.5 Å backbone deviations although they all share a tightly knotted disulfide core. This important variability is likely to arise from the highly diverse loops which connect the successive knotted cysteines. The prediction of structural models for all knottin sequences would open new directions for the analysis of interaction sites and to provide a better understanding of the structural and functional organization of proteins sharing this scaffold. RESULTS: We have designed an automated modeling procedure for predicting the three-dimensionnal structure of knottins. The different steps of the homology modeling pipeline were carefully optimized relatively to a test set of knottins with known structures: template selection and alignment, extraction of structural constraints and model building, model evaluation and refinement. After optimization, the accuracy of predicted models was shown to lie between 1.50 and 1.96 Å from native structures at 50% and 10% maximum sequence identity levels, respectively. These average model deviations represent an improvement varying between 0.74 and 1.17 Å over a basic homology modeling derived from a unique template. A database of 1621 structural models for all known knottin sequences was generated and is freely accessible from our web server at http://knottin.cbs.cnrs.fr. Models can also be interactively constructed from any knottin sequence using the structure prediction module Knoter1D3D available from our protein analysis toolkit PAT at http://pat.cbs.cnrs.fr. CONCLUSIONS: This work explores different directions for a systematic homology modeling of a diverse family of protein sequences. In particular, we have shown that the accuracy of the models constructed at a low level of sequence identity can be improved by 1) a careful optimization of the modeling procedure, 2) the combination of multiple structural templates and 3) the use of conserved structural features as modeling restraints.

BACKGROUND: Structure-dependent substitution matrices increase the accuracy of sequence alignments when the 3D structure of one sequence is known, and are successful e.g. in fold recognition. We propose a new automated method, EvDTree, based on a decision tree algorithm, for automatic derivation of amino acid substitution probabilities from a set of sequence-structure alignments. The main advantage over other approaches is an unbiased automatic selection of the most informative structural descriptors and associated values or thresholds. This feature allows automatic derivation of structure-dependent substitution scores for any specific set of structures, without the need to empirically determine best descriptors and parameters. RESULTS: Decision trees for residue substitutions were constructed for each residue type from sequence-structure alignments extracted from the HOMSTRAD database. For each tree cluster, environment-dependent substitution profiles were derived. The resulting structure-dependent substitution scores were assessed using a criterion based on the mean ranking of observed substitution among all possible substitutions and in sequence-structure alignments. The automatically built EvDTree substitution scores provide significantly better results than conventional matrices and similar or slightly better results than other structure-dependent matrices. EvDTree has been applied to small disulfide-rich proteins as a test case to automatically derive specific substitutions scores providing better results than non-specific substitution scores. Analyses of the decision tree classifications provide useful information on the relative importance of different structural descriptors. CONCLUSIONS: We propose a fully automatic method for the classification of structural environments and inference of structure-dependent substitution profiles. We show that this approach is more accurate than existing methods for various applications. The easy adaptation of EvDTree to any specific data set opens the way for class-specific structure-dependent substitution scores which can be used in threading-based remote homology searches.

MOTIVATION: Genome sequencing projects require the periodic application of analysis tools that can classify and multiply align related protein sequence domains. Full automation of this task requires an efficient integration of similarity and alignment techniques. RESULTS: We have developed a fully automated process that classifies entire protein sequence databases, resulting in alignment of the homologous sequences. The successive steps of the procedure are based on compositional and local sequence similarity searches followed by multiple sequence alignments. Global similarities are detected from the pairwise comparison of amino acid and dipeptide compositions of each protein. After the elimination of all but one sequence from each detected cluster of closely related proteins, the remaining sequences are compiled in a suffix tree which is self-compared to detect local sequence similarities. Sets of proteins which share similar sequence segments are then weighted according to their closeness and multiply aligned using a fast hierarchical dynamic programming algorithm. Computational strategies were devised to minimize computer processing time and memory space requirements. The accuracy of the sequence classifications has been evaluated for 12 462 primary structures distributed over 341 known families. The percentage of sequences with missed or incorrect family assignments was 6.8% on the test set. This low error level is only twice that of the manually constructed PROSITE database ( 3.4% ) and is substantially better than that found for the automatically built PRODOM database ( 34.9% ). AVAILABILITY: The resulting database, called DOMO, is available through database search routine SRS at Infobiogen (http://www.infobiogen.fr/srs5/), EBI (http://srs.ebi.ac.uk:5000/) and EMBL (http://www.embl-heidelberg.de/srs5/) World Wide Web sites. CONTACT: [email protected]

MOTIVATION: Decomposing each protein into modular domains is a basic prerequisite to classify accurately structural units in biological molecules. Boundaries between domains are indicated by two similar amino acid sequence segments located within the same protein (repeats) or within homologous proteins at notably different distances from their respective N- or C-termini. RESULTS: We have developed an automated method that combines such positional constraints derived from various detected pairwise sequence similarities to delineate the modular organization of proteins. The procedure has been applied to a non-redundant data set of 26 990 proteins whose sequences were taken from the PIR and SWISS-PROT databanks and shared <60% sequence identity amongst pairs. The resultant clustering, delineation and multiple alignment of 24 380 sequence fragments yielded a new database of 4364 domain families. Comparison of the domain collection with that of PRODOM indicates a clear improvement in the number and size of domain families, domain boundaries and multiple sequence alignments. The accuracy and sensitivity of the method are illustrated by results obtained for ankyrin-like repeats and EGF-like modules. AVAILABILITY: The resulting database, called DOMO, is available through the database search routine SRS at Infobiogen (http://www.infobiogen.fr/srs5/), EBI (http://srs.ebi.ac.uk:5000/) and EMBL (http://www.embl-heidelberg.de/srs5/) World Wide Web sites. CONTACT: [email protected]