Umeå University's logo

The FISH server is highly accurate in identifying the family membership of domains in a query protein sequence, even in the case of very low sequence identities to known homologues. A performance test using SCOP sequences and an E-value cut-off of 0.1 showed that 99.3% of the top hits are to the correct family saHMM. Matches to a query sequence provide the user not only with an annotation of the identified domains and hence a hint to their function, but also with probable 2D and 3D structures, as well as with pairwise and multiple sequence alignments to homologues with low sequence identity. In addition, the FISH server allows users to upload and search their own protein sequence collection or to quarry public protein sequence data bases with individual saHMMs. The FISH server can be accessed at http://babel.ucmp.umu.se/fish/.

At the core of the FISH (Family Identification with Structure anchored Hidden Markov models, saHMMs) server lies the midnight ASTRAL set. It is a collection of protein domains with low mutual sequence identity within homologous families, according to the structural classification of proteins, SCOP. Here, we evaluate two algorithms for creating the midnight ASTRAL set. The algorithm that limits the number of structural comparisons is about an order of magnitude faster than the all-against-all algorithm. We therefore choose the faster algorithm, although it produces slightly fewer domains in the set. We use the midnight ASTRAL set to construct the structure-anchored Hidden Markov Model data base, saHMM-db, where each saHMM represents one family. Sequence searches using saHMMs provide information about protein function, domain organization, the probable 2D and 3D structure, and can lead to the discovery of homologous domains in remotely related sequences.

MOTIVATION: Order and Disorder prediction using Conditional Random Fields (OnD-CRF) is a new method for accurately predicting the transition between structured and mobile or disordered regions in proteins. OnD-CRF applies CRFs relying on features which are generated from the amino acids sequence and from secondary structure prediction. Benchmarking results based on CASP7 targets, and evaluation with respect to several CASP criteria, rank the OnD-CRF model highest among the fully automatic server group. AVAILABILITY: http://babel.ucmp.umu.se/ond-crf/

Accurate prediction of the change in protein stability due to single amino acid mutations is important for guiding site-directed mutagenesis and other protein-engineering techniques. Recently, different two state predictors became available aimed at predicting if point mutations stabilize or destabilize a protein. Considering the experimental errors and tolerances of protein with respect to mutations, we realized that the neutral mutations, which only slightly affect the protein’s stability, must be considered as well. Here, we present a new classification scheme for a three-state predictor (destabilizing, neutral and stabilizing mutations) based on multi-class support vector machines (SVM). We have created a refined training dataset of single amino acid mutations and evaluate the predictive ability of models trained on homology clustered and non-clustered training data using two different cross validation procedures. The experimental results reveal the significant difference of prediction accuracy according to different evaluation procedures. Furthermore we demonstrate that, for non-clustered model, the prediction accuracy based on the protein sequence information alone is comparable to the prediction accuracy based on protein structure information. On the other hand, for clustered model, the prediction ability is significantly improved when protein tertiary structure information is included. The comparison of prediction accuracy for the two models reveals that the prediction accuracy of mutation stability on clustered proteins is still a challenging task. Moreover, benchmarking by using previously published datasets, demonstrate that our method has an improved prediction performance over many established methods.