Prediction of β-barrel membrane proteins by searching for restricted domains
© Mirus and Schleiff; licensee BioMed Central Ltd. 2005
Received: 20 June 2005
Accepted: 14 October 2005
Published: 14 October 2005
The identification of β-barrel membrane proteins out of a genomic/proteomic background is one of the rapidly developing fields in bioinformatics. Our main goal is the prediction of such proteins in genome/proteome wide analyses.
For the prediction of β-barrel membrane proteins within prokaryotic proteomes a set of parameters was developed. We have focused on a procedure with a low false positive rate beside a procedure with lowest false prediction rate to obtain a high certainty for the predicted sequences. We demonstrate that the discrimination between β-barrel membrane proteins and other proteins is improved by analyzing a length limited region. The developed set of parameters is applied to the proteome of E. coli and the results are compared to four other described procedures.
Analyzing the β-barrel membrane proteins revealed the presence of a defined membrane inserted β-barrel region. This information can now be used to refine other prediction programs as well. So far, all tested programs fail to predict outer membrane proteins in the proteome of the prokaryote E. coli with high reliability. However, the reliability of the prediction is improved significantly by a combinatory approach of several programs. The consequences and usability of the developed scores are discussed.
Genomes of numerous organisms are sequenced. Computer-assisted assignment of coding regions of the organism of interest is the first important step for the understanding of the complex proteomic network . Even though the quality of such predictions will be satisfying in future, the knowledge of the sequences of the gene products alone will not provide insight into their function or localization in the cell. In addition, the emphasis has switched from the study of individual molecules to a large-scale, high-throughput examination of genes and gene products of an organism with the aim of assigning their functions  and placing them into the complex biochemical networks. One kind of information comes from the structural classification of gene products. Since genome and proteome projects result in a rapid increase of information, the biochemical analysis has to be accomplished by in silico predictions . One of the central questions is the localization of proteins since up to 50% of the proteins of a cell have to traverse at least one membrane in order to reach their place of function within organellar compartments . In the past, several prediction programs have been developed for this purpose . However, the analysis of the intracellular localization of a protein is not only limited to the question to which organelle the protein is targeted. One important functional aspect is the distribution of the protein within this cell compartment. For some sub-organellar compartments like thylakoids predictions can be performed based on the targeting signal .
However, to date no differentiation of the signal is found for most sub-organellar localizations. So far, various approaches exist to identify helical transmembrane proteins [7, 8]. More recently, however, the focus was shifted slightly to include the prediction of β-barrel membrane proteins. Initially, structure prediction was applied with reasonable success when proteins already known to form β-barrel structures were modeled . Now, four alternative directions are used in order to newly identify β-barrel proteins out of a genomic/proteomic data set. In the first approach, sequence profile based HMMs for predicting β-barrel membrane proteins were developed [10–12]. The second methodology is based on the alternating hydrophobicity as a measure for β-barrel transmembrane segments . Thirdly, the structural data of the β-barrel membrane proteins were statistically analyzed and certain criteria developed for a linear prediction [14, 15]. The fourth methodology is based on a modified k-nearest neighbor algorithm of the whole sequence amino acid composition [16, 17]. Recently, the combination of several independent procedures for β-barrel membrane protein prediction [18, 19] or their combination with other procedures, e.g. signal sequence prediction [15, 19], was employed to improve the prediction quality.
To evaluate the performance of the developed procedures, test pools are commonly used to derive parameters that discriminate proteins of interest from those of structurally different classes. To avoid an overrepresentation of certain protein families, sequences are removed until each pair of proteins in the pool shares a degree of identity below a certain user defined threshold. Several algorithms have been published to solve this global optimization problem [e.g. [20–22]]. Based on such test pools a comparison of the above mentioned strategies revealed a differential behavior. For example, Deng and co-workers  demonstrated that the linear predictor has a very low false positive but a high false negative rate. In contrast, a broader comparison of the predictors performed by Bagos and co-workers  manifested that the different predictors perform with a similar quality of about 25% false prediction.
We now improved the linear prediction by implementing new parameters and alterations of the previously established parameters based on test pools to increase the reliability and to avoid manual selection. Here, our main goals were to maintain a very low false positive rate and to reduce the high false negative rate of about 51% as reported by Deng and co-workers  for the original prediction method by Wimley . We present parameters for β-barrel membrane protein identification and their prediction performance on the proteome of the prokaryote E. coli.
Summary of the parameters for β-barrel prediction
definition of the parameterb
Number of individual peaks
HPS cut off
HPS cut off
EBSS cut off for strands
Preceding parameters for selection
The new parameter set for linear prediction
Control of the incorporated β-strand number
One of the selection criteria for membrane β-barrel proteins was based on the β-strand number (BSN) of the proteins . The previous BSN was calculated by selecting each individual region with EBSS values above 2.0. Hence, in a stretch of 10 amino acids considered as β-strand several counts can exist if values above 2.0 are separated by values equal or less than 2.0. We changed this algorithm as follows. The first predicted strand now starts at the amino acid with the highest EBSS. The preceding and succeeding nine amino acids are excluded from the search for the highest EBSS in the remaining values to account as starting amino acid of another strand. The β-strand selection procedure stops when no EBSS above 2.0 is left or all amino acids have been assigned to β-strand or pre-β-strand regions.
To establish the selection criteria, we analyzed several test pools (described in Methods). First, the percentage of selected sequences from the sequence pools containing non-barrel proteins (Fig. 2C,D, black lines) with different BSN cut offs in relation to the BSN/aa cut off was determined. For these proteins, a BSN selection cut off of 10 in combination with a BSN/aa cut off of 0.026 results in a 0% false positive selection (Fig. 2C,D). This corresponds to one peak of the EBSS above 2.0 every 40 amino acids, which is above the calculated statistical expected frequency (see above). Previously, the existence of at least one transmembrane α-helix per 100 amino acids was defined as cut off for helical transporters . Comparing the length of the β-strand (10 amino acids on average) and of the membrane inserted helix (20–24 amino acids) as well as the number of the inserted membrane segments (statistically membrane β-barrels contain at least twice as many membrane inserted segments compared to helical transporters) supports the defined BSN/aa score of 0.026. This set of parameters leads to a selection of all sequences in the PDB pool of β-barrel transmembrane proteins (Fig. 2C, grey). For the sequences in the PSort pool of OM proteins a false negative rate of at least 43% is achieved (Fig. 2D, grey dotted). However, the further selection by BBS reduces the false negative rate as described below.
We next analyzed whether the new algorithm for BSN calculation could be used for the generation of topological models of the analyzed proteins. As already discussed above, an over prediction of strands is obtained, especially for larger proteins (Fig. 2E). A detailed analysis of the strands predicted (Fig. 2F) revealed a 68% identical positioning allowing one amino acid mismatch and 80% overlapping positioning of the strands requiring at least five amino acids overlap between structural determined and predicted strands. However, the rate of false positioned strands (false negative and false positive selected strands) is 45%. This analysis suggests that the positioning is not as much the problem as the over prediction and the prediction should be combined with the analysis of other physicochemical parameters.
Development of a new criterion based on the localization of the pore-forming domain
In vivo, a ratio between soluble and helical membrane proteins of at least 10:1 is expected . It is further reasonable to assume that cells do not contain more β-barrel membrane proteins compared to helical membrane proteins. Indeed, other publications discuss a β-barrel membrane protein content within the entire proteome of 2 to 4% [12, 14]. However, our pools represent a ratio of 3.8:1 of NOM (non-outer membrane) to OM proteins. To match the proteomic situation, the false positive prediction rate of the NOM proteins was weighted three times higher than the false negative prediction rate of the OM proteins (Fig. 3B). Hence, a region with lowest false prediction in windows of 250 to 375 amino acids and a BBS-x score of 0.6 to 1.0 could be obtained (Fig. 3B). The lowest false prediction rate was achieved utilizing a window of 275 amino acids with a cut off value of 0.8. Therefore, for the subsequent analysis the BBS is replaced by the BBS in a 275 amino acid window (BBS275). Analysis of the score performance when applied to sequences from E. coli (Fig. 3D) shows, that about 70% of all sequences have a smaller BBS275 compared to the old BBS even though the highest value obtained for BBS275 of E. coli sequences is similar to the highest BBS value (not shown).
Guided by the development and performance of the new BSN score, we developed and analyzed a new score taking into account the alternating hydrophobicity for each predicted strand. Here, for each predicted transmembrane β-strand its alternating hydrophobicity according to equation 1 (E1) was calculated and multiplied with its EBSS value (E2). The final score BSHS (β-strand based hydrophobicity score) is calculated according to equation 3 (E3). The analysis of the performance of the score was performed as described for BBS275. Here, we identified a window of 225 amino acids as best performing (Fig. 3C). This is in line with a homo-oligomeric complex formation of most of the β-barrel membrane proteins, since strands on the protein-protein interface do not necessarily reveal an alternating hydrophobicity as the strands involved in complex formation are not exposed to the membrane lipids .
Development of scores for the linear predictor
After development of three scores for the linear predictor (BSN, BSHS225 and BBS275) we went on to establish selection procedures. They include the three discussed preceding steps by size, TMHMM  prediction and score analysis as discussed above.
However, a 0% false positive predictor does not perform with a low false negative prediction rate. We therefore went on to derive scores for the lowest false prediction rate as well. Since the BSN algorithm leads to an over-prediction of strands, we considered the BSN:BSN/aa selection as an initial step and did not alter the cut off values for selection. Subsequently, the selection by the two scores was performed individually by each score (OR selection) or in combination of both scores (AND selection). Again, we weighted the false positive rate of the NOM pools three times higher as the false negative rate of the OM proteins for the discussed reason. Analyzing the selection performance by the individual BBS275 and BSHS225 (Fig. 4B, Table 2) revealed a score cut off combination of 1.35 and 0.11, respectively. BBS275 and BSHS225 in combination (Fig. 4C, Table 2) result in cut off values of 1.35 and 0.04, respectively. For both procedures a false negative rate of 27.5% and a false positive rate of about 1.2% were obtained based on the analyzed test pools.
Comparison of predictors applied to proteome wide prediction
To achieve an impression of the performance quality, we compared our selection with the performance of MCMBB , BOMP , TMB-Hunt [16, 17] and a predictor just based on the global amino acid distribution of β-barrel proteins . MCMBB selects 10% of the E. coli proteome (Fig. 5A, MCMBB, 565 sequences). Application of the pre- or post-selection by TMHMM (see above) revealed only a slight reduction of the selected pool (Fig. 5A, MCMBB*, 530 sequences). For both selections we found a very high false positive rate of about 70% (Fig. 5C). Interestingly, for the predictor based on the global amino acid composition an even higher number of sequences was selected (Fig. 5A, Global), which was not drastically altered when post-screened with TMHMM (not shown). Even though it was estimated that about 30% of all proteins are helical membrane proteins , it is not considered to be likely that more than 10% of all proteins are β-barrel membrane proteins as discussed above. Therefore, these results raise the question, how reliable scores based on prediction performance on test pools are when transferred to proteome wide prediction.
Using BOMP (Fig. 5A) or TMB-Hunt controlled by the E-value (Fig. 5A, TMB-Hunt°) results in a similar pool size compared to the 0% selection established in here. At default settings BOMP selected 2.23% of the E. coli proteome (Fig. 5A, BOMP) with a false positive rate of 26.4% (Fig. 5C, BOMP) and only two proteins with more than one transmembrane helix according to TMHMM prediction (not shown). TMB-Hunt predicted 1.9% of the E. coli proteome as integral outer membrane proteins (Fig. 5A, TMB-Hunt°) with a false positive rate of 24.1% when requiring both a BBTM protein score >0 and an E-value <1 (Fig. 5C, TMB-Hunt°). A post-selection by TMHMM reduced the ratio of predicted sequences to 1.8% (Fig. 5A, TMB-Hunt°*) and the false positive rate to 19.2% (Fig. 5C, TMB-Hunt°*).
However, simply rejecting all sequences of rank 1–3 from the BOMP selection would not reveal the same result as the overlap procedure. Furthermore, analyzing the rejected sequences we found that most of the sequences rejected from the BOMP prediction are indeed non-β-barrel outer membrane proteins (Fig. 6C, bottom, white). Analyzing the overlap between TMB-Hunt and our AND prediction shows that the BB score  does not show any clear preference for rejection (Fig. 6D), whereas all sequences with an E-value above 0.8 were rejected (Fig. 6E). Interestingly, sequences with very low E-values were rejected as well (Fig. 6E). Analysis of the rejected sequences shows that again mostly non-β-barrel proteins are rejected although the amount of β-barrel proteins removed from the selected pool seems to be increased. Finally, we went on to compare the combinatory approach including our predictor with the combinatory approach among the other programs. Again, utilizing MCMBB resulted in a larger number of selected sequences than the combination of TMB-Hunt and BOMP (Fig. 6F, left). However, to our surprise, the false positive rate was not significantly changed in comparison to the individual programs (compare Fig. 6F right and Fig. 5C). This might be explained as all other programs analyze the entire sequence as such, whereas our prediction is based on a defined region of the sequence.
Summarizing, the combination of our procedure with other predictors increased the quality of the performance. However, this improvement is only achieved by a consensus approach of a domain and a full length sequence based predictor.
The aim of the presented work was to develop better tools or rules for the prediction of β-barrel membrane proteins. In a recent proposal  we obtained a significant false prediction of soluble proteins. First, we went on to optimize the developed scores by implementing a new definition of the BSN and a control parameter for this score (BSN/aa, Fig. 2 and Table 2). Further, we analyzed the domain size optimum for β-barrel discrimination (Fig. 3). Here we learned that the best performance was achieved in a window below 300 amino acids. The latter result is in line with the observation that most porins are about 30–35 kDa . Furthermore, for β-barrel proteins of larger size, clustered pore regions were found. For example the structural modeling of FhaC , ShlB  or Toc75 [15, 35] suggests a soluble domain or long loops in the N-terminal region, whereas only the C-terminal portion seems to be involved in pore formation. It might therefore be suggested that an evolutionary prolongation of the membrane β-barrel proteins occurred facilitating the interaction with other proteins or substrates as seen for Toc75 . This result is interesting for the understanding of the evolutionary development of such proteins. It might point to the fact that a minimal structural unit was the starting seed for the development of larger pores as discussed for helical transporters ("hairpin theory") . Finally, we used a combination of an amino acid distribution based score and the theory that membrane facing strands should reveal an alternating hydrophobicity and calculated a combined score in a 225 amino acid window (BSHS). The window size might reflect that strands involved in homo-oligomerisation do not contain as many hydrophobic amino acids compared to those facing the exterior.
By visual inspection of the structures we determined the average sizes of the continuous region exposed to the lipid membrane and of the region containing the β-barrel. The obtained sizes are ~275 and ~325 amino acids on average, respectively. This corresponds quite well to the window sizes determined for BSHS225 and BBS275. Theoretically, the smallest possible β-barrel membrane domain, an 8-stranded β-barrel of about 80 aa length, should represent the optimal screening window size. But as we are not analyzing each protein separately but a whole pool of proteins, also the larger β-barrels – mostly assembled into homo-oligomeric complexes – have their influence. Here, three major factors contribute to the window sizes determined for BSHS225 and BBS275: (i) The β-barrel has a N-terminal and/or C-terminal extension, (ii) one or more long loops break the compact β-barrel domain into two or more parts and (iii) in homo-oligomeric complexes certain parts of the β-barrel domain are involved into protein-protein binding and therefore do not necessarily show an alternating hydrophobicity which results in a smaller scanning window for the BSHS225 compared to the BBS275. Remarkably, according to Wallin and von Heijne  most of the in there investigated proteins of eubacterial organisms have a local maximum at six transmembrane helices within a segment of about 225 to 275 residues. The average domain sizes of β-barrel and above mentioned helical membrane proteins lie within the same range. Therefore, the best discrimination between the two structurally different classes might be possible within this domain. This finding further supports our approach to identify β-barrel membrane proteins by searching for the transmembrane domain only.
Subsequently, scores for β-barrel membrane protein prediction were developed using test pools (Fig. 4, Table 2) and three preceding rules. First, a selected protein has to be larger than 80 amino acids, since the smallest monomeric transmembrane β-barrel structure consists of 8 strands . Second, if more than one transmembrane helix is identified by TMHMM, the current protein is not considered as a transmembrane β-barrel protein (Fig. 1) and finally, all scores calculated for a sequence have to be larger than zero, regardless of a performed individual or combined selection. In comparison to the previous procedure  we achieved a significant increase of the prediction performance of the E. coli proteome (Fig. 5). Certainly, a factor contributing to this achievement was the greater flexibility of the HPS calculation. Wimley  originally set the loop length to a minimum of four amino acids. By this slight simplification, as Deng and co-workers  also noticed, some hairpins might be missed, because about 28% of the loops are up to three amino acids short . Thus, we kept the window of 25 amino acids for the HPS calculation, but searched for the start of the second β-strand from position 11 to 25, thereby allowing a loop length of 0 to up to 14 amino acids. However, the discriminative power of the linear predictor is limited by the availability of crystal structures of β-barrel membrane proteins. Although about 20 non-redundant structures of this type are currently available in the PDB, they only represent a few families of the diverse group of β-barrel membrane proteins. For example, the important family of β-barrel shaped polypeptide transporters  is still missing. A crystal structure of a member of this family would certainly help to improve the predictive power.
In terms of the prediction performance on the sequence pools we have met our goal of reducing the high false negative rate reported by Deng and co-workers  for Wimley's  original method. Deng and co-workers  developed a HMM for discriminating β-barrel membrane proteins. For screening proteomes they raised the threshold score in order to increase the chance of true positive hits. For our procedures we included in the development of the prediction parameters an optimization for a proteome wide scan by taking care of the proposed in vivo ratio of OM proteins to NOM proteins. Thus, a direct comparison of the performance on proteomes regarding the test pool derived parameters is not possible. This raises the question, if test pools alone are sufficient to receive an impression of the prediction performance on real proteomes. Regarding the generation of test pools not only a broad and diverse collection of proteins but especially the algorithm to reduce the redundancy of the gathered sequences is of central importance. To keep or not to keep a protein – this is here the question. However, there is possibly still a need for improvement of such redundancy removal algorithms. As a consequence, we suggest testing β-barrel membrane protein prediction procedures also on a real proteome. The very well annotated proteome of the prokaryote E. coli  is a good candidate for such a model proteome. This additional testing gives the user a better impression of the reliability of the predictions for prokaryotic proteomes and would allow a better comparison of the scores developed.
The combination of different independent procedures for β-barrel membrane protein prediction [18, 19] was employed to improve the prediction quality. In here we have analyzed and compared several programs and program combinations. These programs can be classified according to their training sets, to the mathematical procedure taken as basis for the prediction or the size restriction for the sequence analysis window. Therefore, the combination of these programs could be achieved based on the difference of one of the named properties. However, we found that predicting sequences with programs differing in the size restriction for the sequence analysis window revealed the lowest false positive rate based on the E. coli proteome. We therefore speculate that the prediction of β-barrel membrane proteins could be further improved employing knowledge based limitations toward the domains, which have to be identified, and global selection approaches in combination.
Test pool generation
In order to evaluate the prediction, the following sequence pools were created with a redundancy of maximal 50%. From the TMPDB  databank we retrieved the file TMPDB_alpha_nr_PR.dat  which contains a set of α-helical transmembrane proteins. We further collected all OM proteins from the experimentally verified ePSortdb dataset v2.0  and removed proteins that are clearly no integral OM proteins and proteins marked as hypothetical. From the same databank all available proteins of the cytoplasmic membrane and the cytoplasm of Gram-negative bacteria were downloaded.
From PDB [43, 44] (version 01/11/2005) we retrieved globular proteins. By SCOP  classification we downloaded from PDB proteins with transmembrane helices and all available transmembrane β-barrels.
As mentioned above, we removed all proteins with more than one transmembrane helix predicted by TMHMM and with less than 80 amino acids. Of the initially 1.235 proteins, 782 survived these steps.
All proteins that are not β-barrel membrane proteins and are not from PDB were accumulated in one sequence pool. They are referred to as PSort NOM protein pool. The OM proteins of PSort and PDB were kept each in separate pools.
For testing our in here developed procedure on a real proteome, the genomic derived sequences deposited at  (from 01/07/2005) for E. coli were used.
Definition of the scores
The algorithms for EBSS, HPS, BBS and BSN were previously described [14, 15]. In brief: The EBSS gives the TM beta-strand probability within a 10 aa sliding window which corresponds to the average length of a TM beta-strand . Approximately 6 aa are required to cross the hydrophobic core of the membrane and further 3 aa for the lipid head group regions of each membrane leaflet . Collecting crystal structures of all β-barrel membrane proteins available in 2002 at a maximum sequence identity of 50% Wimley  calculated the statistical occurrence of amino acids belonging to TM β-strands in the above mentioned membrane regions further differentiating between amino acids exposed to the membrane or oriented towards the interior of the pore. Taking into account the typical β-barrel architecture, the HPS is derived by applying a sliding window calculation to the EBSS values. The HPS is calculated by summing up the greatest EBSS of the first 10 and the following 15 residues. The BBS is calculated by adding all HPS values above 6 normalized to the amino acid number of the sequence.
Calculation of the scores
The BSHS value is derived by calculating the individual score for the β-strand starting at amino acid z. The β-strand position was assigned by the in here redesigned BSN algorithm.
using the water/octanol transfer free energy scale , multiplying the EBSS value
Ystrand(aa=z) = |Xstrand (aa=z)| * EBSS(aa=z) (E2)
adding all values in a defined amino acid window (w) and normalize to that window
exact β-strand score
β-strand based hydrophobicity score
Hidden Markov Model
Markov Chain Model
Special thanks to L. Eichacker for helpful discussions. This work was supported by grants to E.S. from the Deutsche Forschungsgemeinschaft (SFB-TR01), from the Fonds der Chemischen Industrie and from the Volkswagenstiftung.
- Ashurst JL, Collins JE: Gene annotation: prediction and testing. Annu Rev Genomics Hum Genet 2003, 4: 69–88. 10.1146/annurev.genom.4.070802.110300View ArticlePubMedGoogle Scholar
- Drewes G, Bouwmeester T: Global approaches to protein-protein interactions. Curr Opin Cell Biol 2003, 15: 199–205. 10.1016/S0955-0674(03)00005-XView ArticlePubMedGoogle Scholar
- Gerstein M, Hegyi H: Comparing genomes in terms of protein structure: surveys of a finite parts list. FEMS Microbiol Rev 1998, 22: 277–304. 10.1016/S0168-6445(98)00019-9View ArticlePubMedGoogle Scholar
- Schatz G, Dobberstein B: Common principles of protein translocation across membranes. Science 1996, 271: 1519–1526.View ArticlePubMedGoogle Scholar
- Emanuelsson O, von Heijne G: Prediction of organellar targeting signals. Biochim Biophys Acta 2001, 1541: 114–119. 10.1016/S0167-4889(01)00145-8View ArticlePubMedGoogle Scholar
- Westerlund I, von Heijne G, Emanuelsson O: LumenP – a neural network predictor for protein localization in the thylakoid lumen. Protein Sci 2003, 12: 2360–2366. 10.1110/ps.0306003PubMed CentralView ArticlePubMedGoogle Scholar
- Möller S, Croning MD, Apweiler R: Evaluation of methods for the prediction of membrane spanning regions. Bioinformatics 2001, 17: 646–653. 10.1093/bioinformatics/17.7.646View ArticlePubMedGoogle Scholar
- Melen K, Krogh A, von Heijne G: Reliability measures for membrane protein topology prediction algorithms. J Mol Biol 2003, 327: 735–744. 10.1016/S0022-2836(03)00182-7View ArticlePubMedGoogle Scholar
- Jacoboni I, Martelli PL, Fariselli P, De Pinto V, Casadio R: Prediction of the transmembrane regions of beta-barrel membrane proteins with a neural network-based predictor. Protein Sci 2001, 10: 779–787. 10.1110/ps.37201PubMed CentralView ArticlePubMedGoogle Scholar
- Martelli PL, Fariselli P, Krogh A, Casadio R: A sequence-profile-based HMM for predicting and discriminating beta barrel membrane proteins. Bioinformatics 2002, 18(Suppl 1):46–53.View ArticleGoogle Scholar
- Bigelow HR, Petrey DS, Liu J, Przybylski D, Rost B: Predicting transmembrane beta-barrels in proteomes. Nucleic Acids Res 2004, 32: 2566–2577. 10.1093/nar/gkh580PubMed CentralView ArticlePubMedGoogle Scholar
- Bagos PG, Liakopoulos TD, Spyropoulos IC, Hamdrakas SJ: A Hidden Markov Model method, capable of predicting and discriminating β-barrel outer membrane proteins. BMC Bioinformatics 2004, 5: 29. 10.1186/1471-2105-5-29PubMed CentralView ArticlePubMedGoogle Scholar
- Zhai Y, Saier MH Jr: The beta-barrel finder (BBF) program, allowing identification of outer membrane beta-barrel proteins encoded within prokaryotic genomes. Protein Sci 2002, 11: 2196–2207. 10.1110/ps.0209002PubMed CentralView ArticlePubMedGoogle Scholar
- Wimley WC: Toward genomic identification of beta-barrel membrane proteins: composition and architecture of known structures. Protein Sci 2002, 11: 301–312. 10.1110/ps.29402PubMed CentralView ArticlePubMedGoogle Scholar
- Schleiff E, Eichacker LA, Eckart K, Becker T, Mirus O, Stahl T, Soll J: Prediction of the plant beta-barrel proteome: a case study of the chloroplast outer envelope. Protein Sci 2003, 12: 748–759. 10.1110/ps.0237503PubMed CentralView ArticlePubMedGoogle Scholar
- Garrow AG, Agnew A, Westhead DR: TMB-Hunt: a web server to screen sequence sets for transmembrane beta-barrel proteins. Nucleic Acids Res 2005, 33: W188–192. 10.1093/nar/gki384PubMed CentralView ArticlePubMedGoogle Scholar
- Garrow AG, Agnew A, Westhead DR: TMB-Hunt: an amino acid composition based method to screen proteomes for beta-barrel transmembrane proteins. BMC Bioinformatics 2005, 6: 56. 10.1186/1471-2105-6-56PubMed CentralView ArticlePubMedGoogle Scholar
- Moslavac S, Bredemeier R, Mirus O, Granvogl B, Eichacker LA, Schleiff E: Proteomic analysis of the Outer Membrane of Anabaena sp. Strain PCC 7120. J Prot Res 2005, 4: 1330–1338. 10.1021/pr050044cView ArticleGoogle Scholar
- Berven FS, Flikka K, Jensen HB, Eidhammer I: BOMP: a program to predict integral beta-barrel outer membrane proteins encoded within genomes of Gram-negative bacteria. Nucleic Acids Res 2004, 32: W394–399.PubMed CentralView ArticlePubMedGoogle Scholar
- Hobohm U, Scharf M, Schneider R, Sander C: Selection of representative protein data sets. Protein Sci 1992, 1: 409–417.PubMed CentralView ArticlePubMedGoogle Scholar
- Holm L, Sander C: Removing near-neighbour redundancy from large protein sequence collections. Bioinformatics 1998, 14: 423–429. 10.1093/bioinformatics/14.5.423View ArticlePubMedGoogle Scholar
- May ACW: Optimal classification of protein sequences and selection of representative sets from multiple alignments: application to homologous families and lessons for structural genomics. Protein Eng 2001, 14: 209–217. 10.1093/protein/14.4.209View ArticlePubMedGoogle Scholar
- Deng Y, Liu Q, Li YX: Scoring hidden Markov models to discriminate beta-barrel membrane proteins. Comput Biol Chem 2004, 28: 189–194. 10.1016/j.compbiolchem.2004.02.004View ArticlePubMedGoogle Scholar
- Bagos PG, Liakopoulos TD, Promponas VJ, Hamodrakas SJ: Topology prediction of β-barrel outer membrane proteins. PINSA-B 2004, in press.Google Scholar
- Schulz GE: The structure of bacterial outer membrane proteins. Biochim Biophys Acta 2002, 1565: 308–317.View ArticlePubMedGoogle Scholar
- Krogh A, Larsson B, von Heijne G, Sonnhammer EL: Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. J Mol Biol 2001, 305: 567–580. 10.1006/jmbi.2000.4315View ArticlePubMedGoogle Scholar
- Wallin E, von Heijne G: Genome-wide analysis of integral membrane proteins from eubacterial, archaean, and eukaryotic organisms. Protein Sci 1998, 7: 1029–1038.PubMed CentralView ArticlePubMedGoogle Scholar
- Collins RF, Frye SA, Kitmitto A, Ford RC, Tønjum T, Derrick JP: Structure of the Neisseria meningitides Outer Membrane PilQ Secretin Complex at 12 Å Resolution. J Bio Chem 2004, 279: 39750–39756. 10.1074/jbc.M405971200View ArticleGoogle Scholar
- Ferro M, Salvi D, Riviere-Rolland H, Vermat T, Seigneurin-Berny D, Grunwald D, Garin J, Joyard J, Rolland N: Integral membrane proteins of the chloroplast envelope: identification and subcellular localization of new transporters. Proc Natl Acad Sci USA 2002, 99: 11487–11492. 10.1073/pnas.172390399PubMed CentralView ArticlePubMedGoogle Scholar
- Seshadri K, Garemyr R, Wallin E, von Heijne G, Elofsson A: Architecture of beta-barrel membrane proteins: analysis of trimeric porins. Protein Sci 1998, 7: 2026–2032.PubMed CentralView ArticlePubMedGoogle Scholar
- Bagos PG, Liakopoulos TD, Hamodrakas SJ: Finding beta-barrel outer membrane proteins with a markov chain model. WSEAS Transactions on Biology and Biomedicine 2004, 2: 186–189.Google Scholar
- Gromiha MM, Suwa M: A simple statistical method for discriminating outer membrane proteins with better accuracy. Bioinformatics 2005, 21: 961–968. 10.1093/bioinformatics/bti126View ArticlePubMedGoogle Scholar
- Guedin S, Willery E, Tommassen J, Fort E, Drobecq H, Locht C, Jacob-Dubuisson F: Novel topological features of FhaC, the outer membrane transporter involved in the secretion of the Bordetella pertussis filamentous hemagglutinin. J Biol Chem 2000, 275: 30202–30210. 10.1074/jbc.M005515200View ArticlePubMedGoogle Scholar
- Könninger UW, Hobbie S, Benz R, Braun V: The haemolysin-secreting ShlB protein of the outer membrane of Serratia marcescens: determination of surface-exposed residues and formation of ion-permeable pores by ShlB mutants in artificial lipid bilayer membranes. Mol Microbiol 1999, 32: 1212–1225. 10.1046/j.1365-2958.1999.01433.xView ArticlePubMedGoogle Scholar
- Moslavac S, Mirus O, Bredemeier R, Soll J, von Haeseler A, Schleiff E: Conserved pore-forming regions in polypeptide-transporting proteins. FEBS J 2005, 272: 1367–1378. 10.1111/j.1742-4658.2005.04569.xView ArticlePubMedGoogle Scholar
- Ertel F, Mirus O, Bredemeier R, Moslavac S, Becker T, Schleiff E: The Evolutionary related β-barrel polypeptide transporters from P. sativum and Nostoc PCC7120 contain two distinct functional domains. J Biol Chem 2005, 280: 28281–28289. 10.1074/jbc.M503035200View ArticlePubMedGoogle Scholar
- Shimizu T, Mitsuke H, Noto K, Arai M: Internal gene duplication in the evolution of prokaryotic transmembrane proteins. J Mol Biol 2004, 339: 1–15. 10.1016/j.jmb.2004.03.048View ArticlePubMedGoogle Scholar
- Wimley WC: The versatile beta-barrel membrane protein. Curr Opin Struct Biol 2003, 13(4):404–411. 10.1016/S0959-440X(03)00099-XView ArticlePubMedGoogle Scholar
- Mori H: From the Sequence to Cell Modeling: Comprehensive Functional Genomics in Escherichia coli. J Biochem Mol Biol 2004, 37: 83–92.View ArticlePubMedGoogle Scholar
- Ikeda M, Arai M, Okuno T, Shimizu T: TMPDB: a database of experimentally-characterized transmembrane topologies. Nucleic Acids Res 2003, 31: 406–409. 10.1093/nar/gkg020PubMed CentralView ArticlePubMedGoogle Scholar
- TMPDB FTP[ftp://bioinfo.si.hirosaki-u.ac.jp/TMPDB/Release_6.3/]
- Rey S, Acab M, Gardy JL, Laird MR, deFays K, Lambert C, Brinkman FSL: PSORT-DB: A Database of Subcellular Localizations for Bacteria. Nucleic Acids Res 2005, 33: 164–168. 10.1093/nar/gki027View ArticleGoogle Scholar
- Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H, Shindyalov IN, Bourne PE: The Protein Data Bank. Nucleic Acids Res 2000, 28: 235–242. 10.1093/nar/28.1.235PubMed CentralView ArticlePubMedGoogle Scholar
- RCSB Protein Data Bank[http://pdbbeta.rcsb.org/pdb]
- Murzin AG, Brenner SE, Hubbard T, Chothia C: SCOP: a structural classification of proteins database for the investigation of sequences and structures. J Mol Biol 1995, 247: 536–540. 10.1006/jmbi.1995.0159PubMedGoogle Scholar
- NCBI FTP[ftp://ftp.ncbi.nih.gov/genbank/genomes/Bacteria/Escherichia_coli_O157H7]
- White SH, Wimley WC: Hydrophobic interactions of peptides with membrane interfaces. Biochim Biophys Acta 1998, 1376: 339–352.View ArticlePubMedGoogle Scholar
- TMHMM Server v. 2.0[http://www.cbs.dtu.dk/services/TMHMM/]
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.