Predicting absolute contact numbers of native protein structure from amino acid sequence

The contact number of an amino acid residue in a protein structure is defined by the number of C(beta) atoms around the C(beta) atom of the given residue, a quantity similar to, but different from, solvent accessible surface area. We present a method to predict the contact numbers of a protein from its amino acid sequence. The method is based on a simple linear regression scheme and predicts the absolute values of contact numbers. When single sequences are used for both parameter estimation and cross-validation, the present method predicts the contact numbers with a correlation coefficient of 0.555 on average. When multiple sequence alignments are used, the correlation increases to 0.627, which is a significant improvement over previous methods. In terms of discrete states prediction, the accuracies for 2-, 3-, and 10-state predictions are, respectively, 71.4%, 54.1%, and 18.9% with residue type-dependent unbiased thresholds, and 76.3%, 59.2%, and 21.8% with residue type-independent unbiased thresholds. The difference between accessible surface area and contact number from a prediction viewpoint and the application of contact number prediction to three-dimensional structure prediction are discussed.     

Use of a potential of mean force to analyze free energy contributions in protein folding.

A method for calculation of the free energy of residues as a function of residue burial is proposed. The method is based on the potential of mean force, with a reaction coordinate expressed by residue burial. Residue burials are calculated from high-resolution protein structures. The largest individual contributions to the free energy of a residue are found to be due to the hydrophobic interactions of the nonpolar atoms, interactions of the main chain polar atoms, and interactions of the charged groups of residues Arg and Lys. The contribution to the free energy of folding due to the uncharged side chain polar atoms is small. The contribution to the free energy of folding due to the main chain polar atoms is favorable for partially buried residues and less favorable or unfavorable for fully buried residues. Comparison of the accessible surface areas of proteins and model spheres shows that proteins deviate considerably from a spherical shape and that the deviations increase with the size of a protein. The implications of these results for protein folding are also discussed.     

Reconstruction of protein conformations from estimated positions of the C alpha coordinates.

Protein C alpha coordinates are used to accurately reconstruct complete protein backbones and side-chain directions. This work employs potentials of mean force to align semirigid peptide groups around the axes that connect successive C alpha atoms. The algorithm works well for all residue types and secondary structure classes and is stable for imprecise C alpha coordinates. Tests on known protein structures show that root mean square errors in predicted main-chain and C beta coordinates are usually less than 0.3 A. These results are significantly more accurate than can be obtained from competing approaches, such as modeling of backbone conformations from structurally homologous fragments.     

Molecular dissection,tissue localization and Ca2+ binding of the ryanodine receptor of Caenorhabditis elegans

We investigated the possible role of residues at the Ccap position in an alpha-helix on protein stability. A set of 431 protein alpha-helices containing a C'-Gly from the Protein Data Bank (PDB) was analyzed, and the normalized frequencies for finding particular residues at the Ccap position, the average fraction of buried surface area, and the hydrogen bonding patterns of the Ccap residue side-chain were calculated. We found that on average the Ccap position is 70% buried and noted a significant correlation (R=0.8) between the relative burial of this residue and its hydrophobicity as defined by the Gibbs energy of transfer from octanol or cyclohexane to water. Ccap residues with polar side-chains are commonly involved in hydrogen bonding. The hydrogen bonding pattern is such that, the longer side-chains of Glu, Gln, Arg, Lys, His form hydrogen bonds with residues distal (>+/-4) in sequence, while the shorter side-chains of Asp, Asn, Ser, Thr exhibit hydrogen bonds with residues close in sequence (<+/-4), mainly involving backbone atoms. Experimentally we determined the thermodynamic propensities of residues at the Ccap position using the protein ubiquitin as a model system. We observed a large variation in the stability of the ubiquitin variants depending on the nature of the Ccap residue. Furthermore, the measured changes in stability of the ubiquitin variants correlate with the hydrophobicity of the Ccap residue. The experimental results, together with the statistical analysis of protein structures from the PDB, indicate that the key hydrophobic capping interactions between a helical residue (C3 or C4) and a residue outside the helix (C", C3' or C4') are frequently enhanced by the hydrophobic interactions with Ccap residues.     

Statistical potentials for fold assessment

A protein structure model generally needs to be evaluated to assess whether or not it has the correct fold. To improve fold assessment, four types of a residue-level statistical potential were optimized, including distance-dependent, contact, Phi/Psi dihedral angle, and accessible surface statistical potentials. Approximately 10,000 test models with the correct and incorrect folds were built by automated comparative modeling of protein sequences of known structure. The criterion used to discriminate between the correct and incorrect models was the Z-score of the model energy. The performance of a Z-score was determined as a function of many variables in the derivation and use of the corresponding statistical potential. The performance was measured by the fractions of the correctly and incorrectly assessed test models. The most discriminating combination of any one of the four tested potentials is the sum of the normalized distance-dependent and accessible surface potentials. The distance-dependent potential that is optimal for assessing models of all sizes uses both C(alpha) and C(beta) atoms as interaction centers, distinguishes between all 20 standard residue types, has the distance range of 30 A, and is derived and used by taking into account the sequence separation of the interacting atom pairs. The terms for the sequentially local interactions are significantly less informative than those for the sequentially nonlocal interactions. The accessible surface potential that is optimal for assessing models of all sizes uses C(beta) atoms as interaction centers and distinguishes between all 20 standard residue types. The performance of the tested statistical potentials is not likely to improve significantly with an increase in the number of known protein structures used in their derivation. The parameters of fold assessment whose optimal values vary significantly with model size include the size of the known protein structures used to derive the potential and the distance range of the accessible surface potential. Fold assessment by statistical potentials is most difficult for the very small models. This difficulty presents a challenge to fold assessment in large-scale comparative modeling, which produces many small and incomplete models. The results described in this study provide a basis for an optimal use of statistical potentials in fold assessment.     

The histamine N-methyltransferase T105I polymorphism affects active site structure and dynamics

Histamine N-methyltransferase (HNMT) is the primary enzyme responsible for inactivating histamine in the mammalian brain. The human HNMT gene contains a common threonine-isoleucine polymorphism at residue 105, distal from the active site. The 105I variant has decreased activity and lower protein levels than the 105T protein. Crystal structures of both variants have been determined but reveal little regarding how the T105I polymorphism affects activity. We performed molecular dynamics simulations for both 105T and 105I at 37 degrees C to explore the structural and dynamic consequences of the polymorphism. The simulations indicate that replacing Thr with the larger Ile residue leads to greater burial of residue 105 and heightened intramolecular interactions between residue 105 and residues within helix alpha3 and strand beta3. This altered, tighter packing is translated to the active site, resulting in the reorientation of several cosubstrate-binding residues. The simulations also show that the hydrophobic histamine-binding domain in both proteins undergoes a large-scale breathing motion that exposes key catalytic residues and lowers the hydrophobicity of the substrate-binding site.     

Solution structure of the phosphocarrier protein HPr from Bacillus subtilis by two-dimensional NMR spectroscopy.

The solution structure of the phosphocarrier protein, HPr, from Bacillus subtilis has been determined by analysis of two-dimensional (2D) NMR spectra acquired for the unphosphorylated form of the protein. Inverse-detected 2D (1H-15N) heteronuclear multiple quantum correlation nuclear Overhauser effect (HMQC NOESY) and homonuclear Hartmann-Hahn (HOHAHA) spectra utilizing 15N assignments (reported here) as well as previously published 1H assignments were used to identify cross-peaks that are not resolved in 2D homonuclear 1H spectra. Distance constraints derived from NOESY cross-peaks, hydrogen-bonding patterns derived from 1H-2H exchange experiments, and dihedral angle constraints derived from analysis of coupling constants were used for structure calculations using the variable target function algorithm, DIANA. The calculated models were refined by dynamical simulated annealing using the program X-PLOR. The resulting family of structures has a mean backbone rmsd of 0.63 A (N, C alpha, C', O atoms), excluding the segments containing residues 45-59 and 84-88. The structure is comprised of a four-stranded antiparallel beta-sheet with two antiparallel alpha-helices on one side of the sheet. The active-site His 15 residue serves as the N-cap of alpha-helix A, with its N delta 1 atom pointed toward the solvent to accept the phosphoryl group during the phosphotransfer reaction with enzyme I. The existence of a hydrogen bond between the side-chain oxygen atom of Tyr 37 and the amide proton of Ala 56 is suggested, which may account for the observed stabilization of the region that includes the beta-turn comprised of residues 37-40. If the beta alpha beta beta alpha beta (alpha) folding topology of HPr is considered with the peptide chain polarity reversed, the protein fold is identical to that described for another group of beta alpha beta beta alpha beta proteins that include acylphosphatase and the RNA-binding domains of the U1 snRNP A and hnRNP C proteins.     

On the subunit stoichiometry of the F1-ATPase and the sites in it that react specifically with p-fluorosulfonylbenzoyl-5'-adenosine.

During the inactivation of the nucleotide-free F1-ATPase at pH 7.0, by p-fluorosulfonyl[14C]benzoyl-5'-adenosine ([14C]FSBA) in the presence of 20% glycerol, about 4.5 g atoms of 14C are incorporated/350,000 g of enzyme. Isolation of the subunits has shown: (a) over 90% of the incorporated label is associated with the alpha and beta subunits; (b) the amount of label incorporated into the alpha subunit is about 0.5 g atoms/mol which is nonspecifically associated with a number of tyrosine and lysine residues; (c) the amount of radioactivity incorporated into the beta subunit is about 0.9 g atoms/mol which correlates with the degree of inactivation of the enzyme and resides on a single tyrosine residue; (d) up to 2.2 mol of alpha subunit have been isolated from each mole of inactivated enzyme; and (e) about 2 mol of beta subunit have been isolated from each mole of inactivated enzyme. These results account for the incorporation of 4.5 g atoms of 14C which are incorporated/mol of ATPase during inactivation if there are three copies each of the alpha and beta subunit present in the enzyme. It has also been shown that 4-chloro-7-nitrobenzofurazan (NBD-Cl) and FSBA react with different tyrosine residues when they inactivate the ATPase. In addition, it has been shown that the ATPase inactivated with FSBA retains the capacity to bind up to 2.2 mol of [14C]ADP/350,000 g of enzyme.     

The crystal structure of the Epstein-Barr virus protease shows rearrangement of the processed C terminus

Epstein-Barr virus (EBV) belongs to the gamma-herpesvirinae subfamily of the Herpesviridae. The protease domain of the assemblin protein of herpesviruses forms a monomer-dimer equilibrium in solution. The protease domain of EBV was expressed in Escherichia coli and its structure was solved by X-ray crystallography to 2.3A resolution after inhibition with diisopropyl-fluorophosphate (DFP). The overall structure confirms the conservation of the homodimer and its structure throughout the alpha, beta, and gamma-herpesvirinae. The substrate recognition could be modelled using information from the DFP binding, from a crystal contact, suggesting that the substrate forms an antiparallel beta-strand extending strand beta5, and from the comparison with the structure of a peptidomimetic inhibitor bound to cytomegalovirus protease. The long insert between beta-strands 1 and 2, which was disordered in the KSHV protease structure, was found to be ordered in the EBV protease and shows the same conformation as observed for proteases in the alpha and beta-herpesvirus families. In contrast to previous structures, the long loop located between beta-strands 5 and 6 is partially ordered, probably due to DFP inhibition and a crystal contact. It also contributes to substrate recognition. The protease shows a specific recognition of its own C terminus in a binding pocket involving residue Phe210 of the other monomer interacting across the dimer interface. This suggests conformational changes of the protease domain after its release from the assemblin precursor followed by burial of the new C terminus and a possible effect onto the monomer-dimer equilibrium. The importance of the processed C terminus was confirmed using a mutant protease carrying a C-terminal extension and a mutated release site, which shows different solution properties and a strongly reduced enzymatic activity.     

Shape and evolution of thermostable protein structure

Organisms evolved at high temperatures must maintain their proteins' structures in the face of increased thermal disorder. This challenge results in differences in residue utilization and overall structure. Focusing on thermostable/mesostable pairs of homologous structures, we have examined these differences using novel geometric measures: specifically burial depth (distance from the molecular surface to each atom) and travel depth (distance from the convex hull to the molecular surface that avoids the protein interior). These along with common metrics like packing and Wadell Sphericity are used to gain insight into the constraints experienced by thermophiles. Mean travel depth of hyperthermostable proteins is significantly less than that of their mesostable counterparts, indicating smaller, less numerous and less deep pockets. The mean burial depth of hyperthermostable proteins is significantly higher than that of mesostable proteins indicating that they bury more atoms further from the surface. The burial depth can also be tracked on the individual residue level, adding a finer level of detail to the standard exposed surface area analysis. Hyperthermostable proteins for the first time are shown to be more spherical than their mesostable homologues, regardless of when and how they adapted to extreme temperature. Additionally, residue specific burial depth examinations reveal that charged residues stay unburied, most other residues are slightly more buried and Alanine is more significantly buried in hyperthermostable proteins. Proteins 2010. © 2009 Wiley‐Liss, Inc.     

Residue depth: a novel parameter for the analysis of protein structure and stability.

BACKGROUND: Accessible surface area is a parameter that is widely used in analyses of protein structure and stability. Accessible surface area does not, however, distinguish between atoms just below the protein surface and those in the core of the protein. In order to differentiate between such buried residues we describe a computational procedure for calculating the depth of a residue from the protein surface. RESULTS: Residue depth correlates significantly better than accessibility with effects of mutations on protein stability and on protein-protein interactions. The deepest residues in the native state invariably undergo hydrogen exchange by global unfolding of the protein and are often significantly protected in the corresponding molten-globule states. CONCLUSIONS: Depth is often a more useful gage of residue burial than accessibility. This is probably related to the fact that the protein interior and surrounding solvent differ significantly in polarity and packing density. Hence, the strengths of van der Waals and electrostatic interactions between residues in a protein might be expected to depend on the distance of the residue(s) from the protein surface.     

Native atomic burials, supplemented by physically motivated hydrogen bond constraints, contain sufficient information to determine the tertiary structure of small globular proteins

We investigate the possibility that atomic burials, as measured by their distances from the structural geometrical center, contain sufficient information to determine the tertiary structure of globular proteins. We report Monte Carlo simulated annealing results of all-atom hard-sphere models in continuous space for four small proteins: the all-beta WW-domain 1E0L, the alpha/beta protein-G 1IGD, the all-alpha engrailed homeo-domain 1ENH, and the alpha + beta engineered monomeric form of the Cro protein 1ORC. We used as energy function the sum over all atoms, labeled by i, of |R(i) - R(i) (*)|, where R(i) is the atomic distance from the center of coordinates, or central distance, and R(i) (*) is the "ideal" central distance obtained from the native structure. Hydrogen bonds were taken into consideration by the assignment of two ideal distances for backbone atoms forming hydrogen bonds in the native structure depending on the formation of a geometrically defined bond, independently of bond partner. Lowest energy final conformations turned out to be very similar to the native structure for the four proteins under investigation and a strong correlation was observed between energy and distance root mean square deviation (DRMS) from the native in the case of all-beta 1E0L and alpha/beta 1IGD. For all alpha 1ENH and alpha + beta 1ORC the overall correlation between energy and DRMS among final conformations was not as high because some trajectories resulted in high DRMS but low energy final conformations in which alpha-helices adopted a non-native mutual orientation. Comparison between central distances and actual accessible surface areas corroborated the implicit assumption of correlation between these two quantities. The Z-score obtained with this native-centric potential in the discrimination of native 1ORC from a set of random compact structures confirmed that it contains a much smaller amount of native information when compared to a traditional contact Go potential but indicated that simple sequence-dependent burial potentials still need some improvement in order to attain a similar discriminability. Taken together, our results suggest that central distances, in conjunction to physically motivated hydrogen bond constraints, contain sufficient information to determine the native conformation of these small proteins and that a solution to the folding problem for globular proteins could arise from sufficiently accurate burial predictions from sequence followed by minimization of a burial-dependent energy function.     

Benchmarking of TASSER in the ab initio limit

A significant number of protein sequences in a given proteome have no obvious evolutionarily related protein in the database of solved protein structures, the PDB. Under these conditions, ab initio or template-free modeling methods are the sole means of predicting protein structure. To assess its expected performance on proteomes, the TASSER structure prediction algorithm is benchmarked in the ab initio limit on a representative set of 1129 nonhomologous sequences ranging from 40 to 200 residues that cover the PDB at 30% sequence identity and which adopt alpha, alpha + beta, and beta secondary structures. For sequences in the 40-100 (100-200) residue range, as assessed by their root mean square deviation from native, RMSD, the best of the top five ranked models of TASSER has a global fold that is significantly close to the native structure for 25% (16%) of the sequences, and with a correct identification of the structure of the protein core for 59% (36%). In the absence of a native structure, the structural similarity among the top five ranked models is a moderately reliable predictor of folding accuracy. If we classify the sequences according to their secondary structure content, then 64% (36%) of alpha, 43% (24%) of alpha + beta, and 20% (12%) of beta sequences in the 40-100 (100-200) residue range have a significant TM-score (TM-score > or = 0.4). TASSER performs best on helical proteins because there are less secondary structural elements to arrange in a helical protein than in a beta protein of equal length, since the average length of a helix is longer than that of a strand. In addition, helical proteins have shorter loops and dangling tails. If we exclude these flexible fragments, then TASSER has similar accuracy for sequences containing the same number of secondary structural elements, irrespective of whether they are helices and/or strands. Thus, it is the effective configurational entropy of the protein that dictates the average likelihood of correctly arranging the secondary structure elements.     

Secretory processing of the Alzheimer amyloid beta/A4 protein precursor is increased by protein phosphorylation.

The 39-43 residue polypeptide (amyloid beta protein, beta A4) deposited as amyloid in Alzheimer's disease (AD) is derived from a set of 695-770 residue precursors referred to as the amyloid beta A4 protein precursor (beta APP). In each of the 695, 751, and 770 residue precursors, the 43 residue beta A4 is an internal peptide that begins 99 residues from the COOH-terminus of the beta APP. Each holoform is normally cleaved within the beta A4 to produce a large secreted derivative as well as a small membrane associated fragment. Neither of these derivatives can produce amyloid because neither contains the entire beta A4 peptide. In this study, we employ cells stably transfected with full length beta APP695, beta APP751, or beta APP770 expression constructs to show that phorbol ester activation of protein kinase C substantially increases the production of secreted forms from each isoform. By increasing processing of beta APP in the secretory pathway, PKC phosphorylation may help to prevent amyloid deposition.     

Automated amino acid side-chain NMR assignment of proteins using <Superscript>13</Superscript>C- and <Superscript>15</Superscript>N-resolved 3D [<Superscript>1</Superscript>H,<Superscript>1</Superscript>H]-NOESY

ASCAN is a new algorithm for automatic sequence-specific NMR assignment of amino acid side-chains in proteins, which uses as input the primary structure of the protein, chemical shift lists of (1)H(N), (15)N, (13)C(alpha), (13)C(beta) and possibly (1)H(alpha) from the previous polypeptide backbone assignment, and one or several 3D (13)C- or (15)N-resolved [(1)H,(1)H]-NOESY spectra. ASCAN has also been laid out for the use of TOCSY-type data sets as supplementary input. The program assigns new resonances based on comparison of the NMR signals expected from the chemical structure with the experimentally observed NOESY peak patterns. The core parts of the algorithm are a procedure for generating expected peak positions, which is based on variable combinations of assigned and unassigned resonances that arise for the different amino acid types during the assignment procedure, and a corresponding set of acceptance criteria for assignments based on the NMR experiments used. Expected patterns of NOESY cross peaks involving unassigned resonances are generated using the list of previously assigned resonances, and tentative chemical shift values for the unassigned signals taken from the BMRB statistics for globular proteins. Use of this approach with the 101-amino acid residue protein FimD(25-125) resulted in 84% of the hydrogen atoms and their covalently bound heavy atoms being assigned with a correctness rate of 90%. Use of these side-chain assignments as input for automated NOE assignment and structure calculation with the ATNOS/CANDID/DYANA program suite yielded structure bundles of comparable quality, in terms of precision and accuracy of the atomic coordinates, as those of a reference structure determined with interactive assignment procedures. A rationale for the high quality of the ASCAN-based structure determination results from an analysis of the distribution of the assigned side chains, which revealed near-complete assignments in the core of the protein, with most of the incompletely assigned residues located at or near the protein surface.     

1H-NMR studies at 500 MHz of a neutral disaccharide and sulphated di-, tetra-, hexa- and larger oligosaccharides obtained by endo-beta-galactosidase treatment of keratan sulphate

In the preceding paper in this journal, the major oligosaccharides obtained by endo-beta-galactosidase digestion of bovine corneal keratan sulphate were identified as a neutral disaccharide, GlcNAc beta 1-3Gal, and sulphated di-, tetra-, hexa-, octa- and decasaccharides based on the sequence (-3/4GlcNAc beta 1-3Gal beta 1-)n having 1, 3, 5, 7 and 9 sulphate groups, respectively. In the present study, these oligosaccharides have been analysed by 500-MHz 1H-NMR spectroscopy using spin-decoupling and two-dimensional correlated spectroscopy experiments. The NMR data confirm the beta-configuration of all the interglycosidic linkages and are consistent with an alternating sequence of----4GlcNAc and----3Gal, a non-reducing-end N-acetylglucosamine residue and a reducing-end galactose residue. The NMR data have also established that a sulphate group is linked to the C6 position of all sugar residues except the reducing-end galactose as follows: (Formula: see text). The signals of the protons attached to the sulphated carbon atoms show marked downfield shifts (approximately 0.4 ppm from equivalent protons of non-sulphated carbon atoms), while the protons at C5 vicinal to sulphated atoms show a change of 0.1-0.2 ppm and other protons of the sulphated monosaccharides show smaller changes in chemical shift (0.01-0.1 ppm). The proton at C4 of the non-sulphated reducing-end galactose linked at C3 also shows a significant change in chemical shift (0.03 ppm).     

Analysis of six protein structures predicted by comparative modeling techniques

The protein structures of six comparative modeling targets were predicted in a procedure that relied on improved energy minimization, without empirical rules, to position all new atoms. The structures of human nucleoside diphosphate kinase NM23-H2, HPr from Mycoplasma capricolum, 2Fe-2S ferredoxin from Haloarcula marismortui, eosinophil-derived neurotoxin (EDN), mouse cellular retinoic acid protein I (CRABP1), and P450eryf were predicted with root mean square deviations on Cα atoms of 0.69, 0.73, 1.11, 1.48, 1.69, and 1.73 Å, respectively, compared to the target crystal structures. These differences increased as the sequence similarity between the target and parent proteins decreased from about 60 to 20% identity. More residues were predicted than form the common region shared by the two crystal structures. In most cases insertions or deletions between the target and the related protein of known structure were not correctly positioned. One two residue insertion in CRABP1 was predicted in the correct conformation, while a nine residue insertion in EDN was predicted in the correct spatial region, although not in the correct conformation. The positions of common cofactors and their binding sites were predicted correctly, even when overall sequence similarity was low. © 1995 Wiley-Liss, Inc.     

Comprehensive mutagenesis of HIV-1 protease: a computational geometry approach

A computational geometry technique based on Delaunay tessellation of protein structure, represented by C(alpha) atoms, is used to study effects of single residue mutations on sequence-structure compatibility in HIV-1 protease. Profiles of residue scores derived from the four-body statistical potential are constructed for all 1881 mutants of the HIV-1 protease monomer and compared with the profile of the wild-type protein. The profiles for an isolated monomer of HIV-1 protease and the identical monomer in a dimeric state with an inhibitor are analyzed to elucidate changes to structural stability. Protease residues shown to undergo the greatest impact are those forming the dimer interface and flap region, as well as those known to be involved in inhibitor binding.     

COMSAT: Residue contact prediction of transmembrane proteins based on support vector machines and mixed integer linear programming

In this article, we present COMSAT, a hybrid framework for residue contact prediction of transmembrane (TM) proteins, integrating a support vector machine (SVM) method and a mixed integer linear programming (MILP) method. COMSAT consists of two modules: COMSAT_SVM which is trained mainly on position–specific scoring matrix features, and COMSAT_MILP which is an ab initio method based on optimization models. Contacts predicted by the SVM model are ranked by SVM confidence scores, and a threshold is trained to improve the reliability of the predicted contacts. For TM proteins with no contacts above the threshold, COMSAT_MILP is used. The proposed hybrid contact prediction scheme was tested on two independent TM protein sets based on the contact definition of 14 Å between Cα‐Cα atoms. First, using a rigorous leave‐one‐protein‐out cross validation on the training set of 90 TM proteins, an accuracy of 66.8%, a coverage of 12.3%, a specificity of 99.3% and a Matthews' correlation coefficient (MCC) of 0.184 were obtained for residue pairs that are at least six amino acids apart. Second, when tested on a test set of 87 TM proteins, the proposed method showed a prediction accuracy of 64.5%, a coverage of 5.3%, a specificity of 99.4% and a MCC of 0.106. COMSAT shows satisfactory results when compared with 12 other state‐of‐the‐art predictors, and is more robust in terms of prediction accuracy as the length and complexity of TM protein increase. COMSAT is freely accessible at http://hpcc.siat.ac.cn/COMSAT/ . Proteins 2016; 84:332–348. © 2016 Wiley Periodicals, Inc.