Protein subcellular location prediction

The function of a protein is closely correlated with its subcellular location. With the rapid increase in new protein sequences entering into data banks, we are confronted with a challenge: is it possible to utilize a bioinformatic approach to help expedite the determination of protein subcellular locations? To explore this problem, proteins were classified, according to their subcellular locations, into the following 12 groups: (1) chloroplast, (2) cytoplasm, (3) cytoskeleton, (4) endoplasmic reticulum, (5) extracell, (6) Golgi apparatus, (7) lysosome, (8) mitochondria, (9) nucleus, (10) peroxisome, (11) plasma membrane and (12) vacuole. Based on the classification scheme that has covered almost all the organelles and subcellular compartments in an animal or plant cell, a covariant discriminant algorithm was proposed to predict the subcellular location of a query protein according to its amino acid composition. Results obtained through self-consistency, jackknife and independent dataset tests indicated that the rates of correct prediction by the current algorithm are significantly higher than those by the existing methods. It is anticipated that the classification scheme and concept and also the prediction algorithm can expedite the functionality determination of new proteins, which can also be of use in the prioritization of genes and proteins identified by genomic efforts as potential molecular targets for drug design.

Prediction of membrane protein types and subcellular locations

Membrane proteins are classified according to two different schemes. In scheme 1, they are discriminated among the following five types: (1) type I single-pass transmembrane, (2) type II single-pass transmembrane, (3) multipass transmembrane, (4) lipid chain-anchored membrane, and (5) GPI-anchored membrane proteins. In scheme 2, they are discriminated among the following nine locations: (1) chloroplast, (2) endoplasmic reticulum, (3) Golgi apparatus, (4) lysosome, (5) mitochondria, (6) nucleus, (7) peroxisome, (8) plasma, and (9) vacuole. An algorithm is formulated for predicting the type or location of a given membrane protein based on its amino acid composition. The overall rates of correct prediction thus obtained by both self-consistency and jackknife tests, as well as by an independent dataset test, were around 76-81% for the classification of five types, and 66-70% for the classification of nine cellular locations. Furthermore, classification and prediction were also conducted between inner and outer membrane proteins; the corresponding rates thus obtained were 88-91%. These results imply that the types of membrane proteins, as well as their cellular locations and other attributes, are closely correlated with their amino acid composition. It is anticipated that the classification schemes and prediction algorithm can expedite the functionality determination of new proteins. The concept and method can be also useful in the prioritization of genes and proteins identified by genomics efforts as potential molecular targets for drug design.

Using discriminant function for prediction of subcellular location of prokaryotic proteins

The discriminant function algorithm was introduced to predict the subcellular location of proteins in prokaryotic organisms from their amino-acid composition. The rate of correct prediction for the three possible subcellular locations of prokaryotic proteins studied by Reinhardt and Hubbard (Nucleic Acid Research, 1998, 26:2230-2236) was 90% by the self-consistency test, and 87% by the jackknife test. These rates are considerably higher than the results recently reported by them using the neural network method. Furthermore, the test procedure adopted here is also more rigorous. The core of the current algorithm is the covariance matrix, through which the collective interactions among different amino-acid components of a protein can be reflected. It is anticipated that, owing to the intimate correlation of the function of a protein with its subcellular location, the current algorithm will become a useful tool for the systematic analysis of genome data.

Singular points of protein beta-sheets

Protein beta-sheets can be regarded as surfaces. Two surfaces can be connected along a common edge to form a larger surface, or two edges of a surface can coalesce to form a closed sheet such as a beta-barrel. Singular points are locations where these connections are not perfect. In protein beta-sheets, a singular point is characterized by a residue separating two beta-ladders. In this paper, we study the singular points of protein beta-sheets from the surface topologic viewpoint, summarize our search results from the protein structural data in the Protein Data Bank, and present examples where singular points are near the active sites and may contribute to forming the proper relative positions of catalytic residues.

Artificial neural network method for predicting HIV protease cleavage sites in protein

Knowledge of the polyprotein cleavage sites by HIV protease will refine our understanding of its specificity, and the information thus acquired will be useful for designing specific and efficient HIV protease inhibitors. The search for inhibitors of HIV protease will be greatly expedited if one can find an accurate, robust, and rapid method for predicting the cleavage sites in proteins by HIV protease. In this paper, Kohonen's self-organization model, which uses typical artificial neural networks, is applied to predict the cleavability of oligopeptides by proteases with multiple and extended specificity subsites. We selected HIV-1 protease as the subject of study. We chose 299 oligopeptides for the training set, and another 63 oligopeptides for the test set. Because of its high rate of correct prediction (58/63 = 92.06%) and stronger fault-tolerant ability, the neural network method should be a useful technique for finding effective inhibitors of HIV protease, which is one of the targets in designing potential drugs against AIDS. The principle of the artificial neural network method can also be applied to analyzing the specificity of any multisubsite enzyme.

Domain structural class prediction

The structural class of a protein domain can be approximately predicted according to its amino acid composition. However, can the prediction quality be improved by taking into account the coupling effect among different amino acid components? This question has evoked much controversy because completely different conclusions have been obtained by different investigators. To resolve such a perplexing problem, predictions by means of various algorithms were performed based on the SCOP database (Murzin et aL, 1995), which is more natural and reliable for the study of structural classes because it is based on evolutionary relationships and on the principles that govern their three-dimensional structure. The results obtained using both resubstitution and jackknife tests indicated that the overall rates of correct prediction by an algorithm incorporating the coupling effect among different amino acid components were significantly higher than those by the algorithms that did not include such an effect. A completely consistent conclusion was also obtained when tests were performed on two large independent testing datasets classified into four and seven structural classes, respectively. It is revealed through an analysis that the reasons for reaching the opposite conclusion are mainly due to (1) misclassifying structural classes according to a conceptually incorrect rule, (2) misapplying the component-coupled algorithm by ignoring some important factors and (3) misrepresenting structural classes with statistically insignificant training subsets. Clarification of these problems would be instructive for effectively using the prediction algorithm and correctly interpreting the results.

Prediction of beta-turns

Kohonen's self-organization model, a neural network model, is applied to predict the beta-turns in proteins. There are 455 beta-turn tetrapeptides and 3807 non-beta-turn tetrapeptides in the training database. The rates of correct prediction for the 110 beta-turn tetrapeptides and 30,229 non-beta-turn tetrapeptides in the testing database are 81.8% and 90.7%, respectively. The high quality of prediction of neural network model implies that the residue-coupled effect along a polypeptide chain is important for the formation of reversal turns, such as beta-turns, during the process of protein folding.

Prediction of protein structural classes by modified mahalanobis discriminant algorithm

We first discuss quantitative rules for determining the protein structural classes based on their secondary structures. Then we propose a modification of the least Mahalanobis distance method for prediction of protein classes. It is a generalization of a quadratic discriminant function to the case of degenerate covariance matrices. The resubstitution tests and leave-one-out tests are carried out to compare several methods. When the class sample sizes or the covariance matrices of different classes are significantly different, the modified method should be used to replace the least Mahalanobis distance method. Two lemmas for the derivation of our new algorithm are proved in an appendix.

Prediction and classification of domain structural classes

Can the coupling effect among different amino acid components be used to improve the prediction of protein structural classes? The answer is yes according to the study by Chou and Zhang (Crit. Rev. Biochem. Mol. Biol. 30:275-349, 1995), but a completely opposite conclusion was drawn by Eisenhaber et al. when using a different dataset constructed by themselves (Proteins 25:169-179, 1996). To resolve such a perplexing problem, predictions were performed by various approaches for the datasets from an objective database, the SCOP database (Murzin, Brenner, Hubbard, and Chothia. J. Mol. Biol. 247:536-540, 1995). According to SCOP, the classification of structural classes for protein domains is based on the evolutionary relationship and on the principles that govern the 3D structure of proteins, and hence is more natural and reliable. The results from both resubstitution tests and jackknife tests indicate that the overall rates of correct prediction by the algorithm incorporated with the coupling effect among different amino acid components are significantly higher than those by the algorithms without using such an effect. It is elucidated through an analysis that the main reasons for Eisenhaber et al. to have reached an opposite conclusion are the result of (1) misusing the component-coupled algorithm, and (2) using a conceptually incorrect rule to classify protein structural classes. The formulation and analysis presented in this article are conducive to clarify these problems, helping correctly to apply the prediction algorithm and interpret the results.

Phase II trial of intrapleural paclitaxel injection for non-small-cell lung cancer patients with malignant pleural effusions

A phase II clinical trial of intrapleural paclitaxel injection for malignant effusions of non-small-cell lung cancer (NSCLC) was conducted in order to evaluate the efficacy and toxicity profile of paclitaxel pleurodesis in patients with malignant effusions. From February to May of 1996, 15 NSCLC patients with malignant pleural effusions were enrolled on study. After adequate drainage and assurance of lung re-expansion, paclitaxel 125 mg m-2 diluted in normal saline was infused through a preinserted pig-tail catheter which was removed 2 h later. Chest radiography and sonography were scheduled 4 days later; depending on whether there remained a significant amount of pleural effusion, further drainage by needle thoracentesis or by a pig-tail catheter was performed. All patients were assessable for toxicity. Ipsilateral chest and/or shoulder pain, fever, facial flushing and nausea were the most frequent side-effects. Grade 4 neutropenia, grade 3 anaemia, and grade 3 renal impairment occurred in one patient each. Fourteen patients were evaluable for response at the end of the fourth week. Overall response rate of pleural effusion in evaluable patients was 92.9%, with a complete response rate of 28.6%. There was one out of 14 evaluable patients whose measurable tumour lesion decreased by more than 50% (partial response). No disease progression was noted among evaluable patients at the end of the fourth week. It is concluded that paclitaxel is a useful agent for the treatment of malignant pleural effusions. Because of its relatively low systemic toxicity, intrapleural paclitaxel injection in combination with systemic chemotherapy or radiotherapy can be considered in treating NSCLC patients with malignant pleural effusions.

A model for structure-dependent binding of Congo red to Alzheimer beta-amyloid fibrils

The cytotoxic A beta fibril is a logical candidate for the entity causing the initiating damage to neurons in Alzheimer's disease and Down's syndrome. We have derived a model of binding for the dye molecule, Congo red (CR), to a beta-sheet structure of beta-amyloid (1-42). This model is based on the crystal coordinates of CR binding to porcine insulin fibrils from Turnell and Finch. Intact insulin is composed of protein dimers and X-ray diffraction studies show that CR intercalates between two insulin monomers at an interface formed by a pair of antiparallel beta-strands. The intercalation of CR has disrupted the four main-chain hydrogen bonds between the two beta-strands, but they are still tethered with each other through new hydrogen bonds with the CR nitrogen atoms. The CR molecule has been aligned along the homologous stretch of amino acids in Alzheimer beta peptide (two molecules in antiparallel distorted or pseudo beta-sheet conformation) using the crystal coordinates from the Turnell-Finch paper to arrive at a putative structure for CR binding to Alzheimer's amyloid fibrils.

Prediction of the tertiary structure and substrate binding site of caspase-8

The caspases represent a family of sulfhydryl proteases that play important regulatory roles in the cell. The tertiary structure of the protease domain of caspase-8, also called FLICE, has been predicted by a segment match modeling procedure. First, the atomic coordinates of the catalytic domain of caspase-3, also called CPP32, a member of the family that is closely related to caspase-8, were determined based upon the crystal structure of human caspase-1 (interleukin converting enzyme). Then, the caspase-3 structure was used as a template for modeling the protease domain of caspase-8. The resulting structure shows the expected level of similarity with the conformations of caspases-1 and -3 for which crystal structures have been determined. Moreover, the subsite contacts between caspase-8 and the covalently linked inhibitor, Ac-DEVD-aldehyde, are only slightly different from those seen in the caspase-3 enzyme/inhibitor complex. The model of caspase-8 can serve as a reference for subsite analysis relative to design of enzyme inhibitors that may find therapeutic application.

Prediction and classification of alpha-turn types

Tight turns play an important role in globular proteins from both the structural and functional points of view. Of tight turns, beta-turns and gamma-turns have been extensively studied, but alpha-turns were little investigated. Recently, a systematic search for alpha-turns was conducted by V. Pavone et al. [(1996) Biopolymers, Vol. 38, pp. 705-721] from 190 proteins (221 protein chains). They found 356 alpha-turns that were classified into nine different types according to their backbone trajectory features. In view of this new discovery, a sequence-coupled model based on Markov chain theory is proposed for predicting the alpha-turn types in proteins. The high rates of correct prediction by resubstitution test and jackknife test imply that that the formation of different alpha-turn types is evidently correlated with the sequence of a pentapeptide, and hence can be approximately predicted based on the sequence information of the pentapeptide alone, although the role of its interaction with the other part of a protein cannot be completely ignored. The algorithm presented here can also be used to conduct the prediction in which a distinction between alpha-turns and non-alpha-turns is also required.

Prediction of the tertiary structure of the complement control protein module

Complement control protein (CCP) modules, or short consensus repeats (SCR), exist in a wide variety of complement and adhesion proteins, principally the selectins. We have predicted the three-dimensional structure of a CCP module based upon secondary structural information derived by two-dimensional NMR [Barlow et al. (1991), Biochemistry 30, 997-1004]. Accordingly, the CCP is predicted to contain seven beta-strands with extensive hydrogen-bonding interactions, and shows a compact, globular structure. Comparison of this model to the X-ray structure of a kringle domain suggests that the CCP unit is more compact than a kringle structure, and that despite their similarities in size and disulfide bond format, the two are not homologous. Although the function of CCP domains is unknown, it is hoped that the structural model presented herein will facilitate further inquiry into how they contribute to so many systems of biological importance.

Artificial neural network method for predicting the specificity of GalNAc-transferase

The specificity of GalNAc-transferase is consistent with the existence of an extended site composed of nine subsites, denoted by R4, R3, R2, R1, R0, R1', R2', R3', and R4', where the acceptor at R0 is either Ser or Thr to which the reducing monosaccharide is anchored. To predict whether a peptide will react with the enzyme to form a Ser- or Thr-conjugated glycopeptide, a neural network method--Kohonen's self-organization model is proposed in this paper. Three hundred five oligopeptides are chosen for the training site, with another 30 oligopeptides for the test set. Because of its high correct prediction rate (26/30 = 86.7%) and stronger fault-tolerant ability, it is expected that the neural network method can be used as a technique for predicting O-glycosylation and designing effective inhibitors of GalNAc-transferase. It might also be useful for targeting drugs to specific sites in the body and for enzyme replacement therapy for the treatment of genetic disorders.

Classification and prediction of beta-turn types

Although a beta-turn consists of only four amino acids, it assumes many different types in proteins. Is this basically dependent on the tetrapeptide sequence alone or is it due to a variety of interactions with the other part of a protein? To answer this question, a residue-coupled model is proposed that can reflect the sequence-coupling effect for a tetrapeptide in not only a beta-turn or non-beta-turn, but also different types of a beta-turn. The predicted results by the model for 6022 tetrapeptides indicate that the rates of correct prediction for beta-turn types I, I', II, II', VI, and VIII and non-beta-turns are 68.54%, 93.60%, 85.19%, 97.75%, 100%, 88.75%, and 61.02%, respectively. Each of these seven rates is significantly higher than 1/7 = 14.29%, the completely randomized rate, implying that the formation of different beta-turn types or non-beta-turns is considerably correlated with the sequences of a tetrapeptide.

Disposition of amphiphilic helices in heteropolar environments

It is known that alpha helices in globular proteins usually consist of two types of residues, hydrophobic and hydrophilic, with the number of each type being roughly equal. Except for many transmembrane helices, alpha-helices are generally amphiphilic to some degree. This is not entirely surprising because alpha-helices typically reside in heteropolar environments that arise from the polar aqueous solution that surrounds a protein and the apolar "hydrophobic core" located at its center. The packing of alpha-helices in such heteropolar environments is driven by the minimization of free energy brought about by placing hydrophobic sidechains into apolar environments and hydrophilic sidechains into polar environments. The interface between the two environments can be characterized by an interfacial plane, called the demarcation plane, that optimally separates the two classes of residues. The inclination angle omega between the axis of the helix and the demarcation plane provides a measure of the degree of amphiphilicity of an alpha-helix. For highly amphiphilic helices, omega approximately 0. The inclination angle provides a new measure of amphiphilicity that complements the hydrophobic moments of Eisenberg et al. Based on the simple physical model described above, an algorithm is developed for predicting the helix inclination angle. The calculated results show that the inclination angle for most alpha-helices extracted from globular proteins is less than 25 degrees in magnitude. This suggests that helices found in globular proteins tend to be reasonably amphiphilic with half their face dominated by hydrophobic residues and the other half by hydrophilic residues. A new two-dimensional representation that characterizes the disposition of hydrophobic and hydrophilic residues in alpha-helices, called a "wenxiang diagram," is presented. The wenxiang diagram can also be used as an important element to represent a protein molecule.

Prediction of beta-turns

A residue-coupled model is proposed to predict the beta-turns in proteins. The rates of correct prediction for the 455 beta-turn tetrapeptides and 3807 non-beta-turn tetrapeptides in the training database are 94.7 and 81.3%, respectively. The rates of correct prediction for the 110 beta-turn tetrapeptides and 30,229 non-beta-turn tetrapeptides in the testing database are 80.0 and 80.2%, respectively. Compared with the rates of correct prediction based on the residue-independent model reported previously, the quality of prediction is significantly improved by the new model, implying that the residue-coupled effect along a polypeptide chain is important for the formation of reversal turns, such as beta-turns, during the process of protein folding.

The benzylthio-pyrimidine U-31,355, a potent inhibitor of HIV-1 reverse transcriptase

U-31,355, or 4-amino-2-(benzylthio)-6-chloropyrimidine is an inhibitor of human immunodeficiency virus type 1 (HIV-1) reverse transcriptase (RT) and possesses anti-HIV activity in HIV-1-infected lymphocytes grown in tissue culture. The compound acts as a specific inhibitor of the RNA-directed DNA polymerase function of HIV-1RT and does not impair the functions of the DNA-catalyzed DNA polymerase or the Rnase H of the enzyme. Kinetic studies were carried out to elucidate the mechanism of RT inhibition by U-31,355. The data were analyzed using Briggs-Haldane kinetics, assuming that the reaction is ordered in that the template:primer binds to the enzyme first, followed by the addition of dNTP, and that the polymerase is a processive enzyme. Based on these assumptions, a velocity equation was derived that allows the calculation of all the essential forward and backward rate constants for the reactions occurring between the enzyme, its substrates, and the inhibitor. The results obtained indicate that U-31,355 acts as a mixed inhibitor with respect to the template:primer and dNTP binding sites associated with the RNA-directed DNA polymerase domain of the enzyme. The inhibitor possessed a significantly higher binding affinity for the enzyme-substrate complexes, than for the free enzyme and consequently did not directly affect the functions of the substrate binding sites. Therefore, U-31,355 appears to impair an event occurring after the formation of the enzyme-substrate complexes, which involves either inhibition of the phosphoester bond formation or translocation of the enzyme relative to its template:primer following the formation of the ester bond. Moreover, the potency of U-31,355 depends on the base composition of the template:primer in that the inhibitor showed a much higher binding affinity for the enzyme-poly (rC):(dG)10 complexes than for the poly (rA):(dT)10 complexes.

Knowledge-based model building of the tertiary structures for lectin domains of the selectin family

A combination of a knowledge-based approach and energy minimization was used to predict the three-dimensional structures of the lectin domains of P-selectin, E-selectin, and L-selectin, respectively. Each of these domains contains 118 amino acids. The starting points for energy minimization were generated based on a framework that consists of a number of separated segments derived from the structure-known carbohydrate-recognition domain of the mannose-binding protein (MBP), which belongs to the same C-type lectin family as the selectin molecules do. The structures thus found for P-, L-, and E-selectin lectin domains share a common feature, i.e., they all contain two alpha-helices, and two antiparallel beta-sheets of which one is formed by two strands (strands 1 and 5) and the other by three (strands 2, 3, and 4). Besides, they all possess two intact disulfide bonds formed by the pair of Cys-19 and Cys-117, and the pair of Cys-90 and Cys-109. The root-mean-square deviations calculated over the set of backbone atoms between P- and L-selectin lectin domains is 3.10 A, that between P- and E-selectin lectin domains 2.48 A, and that between L- and E-selectin lectin domains 3.07 A. A notable feature is the convergence-divergence duality of the 77-107 polypeptide in the three domains; i.e., part of the peptide is folded into a closely similar conformation, and part of it into a highly different one.

Do "antisense proteins" exist?

A DNA double helix consists of two complementary strands antiparallel with each other. One of them is the sense chain, while the other is an antisense chain which does not directly involve the protein-encoding process. The reason that an antisense chain cannot encode for a protein is generally attributed to the lack of certain preconditions such as a promotor and some necessary sequence segments. Suppose it were provided with all these preconditions, could an antisense chain encode for an "antisense protein"? To answer this question, an analysis has been performed based on the existing database. Nine proteins have been found that have a 100% sequence match with the hypothetical antisense proteins derived from the known Escherichia coli antisense chains.