Distinct position-specific sequence features of hexa-peptides that form amyloid-fibrils: application to discriminate between amyloid fibril and amorphous β-aggregate forming peptide sequences
© Thangakani et al.; licensee BioMed Central Ltd. 2013
Published: 9 May 2013
Comparison of short peptides which form amyloid-fibrils with their homologues that may form amorphous β-aggregates but not fibrils, can aid development of novel amyloid-containing nanomaterials with well defined morphologies and characteristics. The knowledge gained from the comparative analysis could also be applied towards identifying potential aggregation prone regions in proteins, which are important for biotechnology applications or have been implicated in neurodegenerative diseases. In this work we have systematically analyzed a set of 139 amyloid-fibril hexa-peptides along with a highly homologous set of 168 hexa-peptides that do not form amyloid fibrils for their position-wise as well as overall amino acid compositions and averages of 49 selected amino acid properties.
Amyloid-fibril forming peptides show distinct preferences and avoidances for amino acid residues to occur at each of the six positions. As expected, the amyloid fibril peptides are also more hydrophobic than non-amyloid peptides. We have used the results of this analysis to develop statistical potential energy values for the 20 amino acid residues to occur at each of the six different positions in the hexa-peptides. The distribution of the potential energy values in 139 amyloid and 168 non-amyloid fibrils are distinct and the amyloid-fibril peptides tend to be more stable (lower total potential energy values) than non-amyloid peptides. The average frequency of occurrence of these peptides with lower than specific cutoff energies at different positions is 72% and 50%, respectively. The potential energy values were used to devise a statistical discriminator to distinguish between amyloid-fibril and non-amyloid peptides. Our method could identify the amyloid-fibril forming hexa-peptides to an accuracy of 89%. On the other hand, the accuracy of identifying non-amyloid peptides was only 54%. Further attempts were made to improve the prediction accuracy via machine learning. This resulted in an overall accuracy of 82.7% with the sensitivity and specificity of 81.3% and 83.9%, respectively, in 10-fold cross-validation method.
Amyloid-fibril forming hexa-peptides show position specific sequence features that are different from those which may form amorphous β-aggregates. These positional preferences are found to be important features for discriminating amyloid-fibril forming peptides from their homologues that don't form amyloid-fibrils.
Outcome of the competition between functionally active and inactive aggregated forms is critical to a protein's fate in vivo and in vitro. Indeed, aggregation is an ancient threat to proper folding of proteins and it must be overcome by proteins from all organisms to maintain their native functional states. The aggregation of endogenous proteins causes several diseases in humans and animals. Aggregation is also a major hurdle in successful development of biopharmaceutical drug products . In converse biotechnology applications, creation of protein and peptide aggregates with well defined morphologies is of interest for development of nano-materials with desired characteristics.
On the computational point of view, amino acid properties such as hydrophobicity, β-strand propensity, charge and solubility of amyloid forming peptides have been analyzed and used to predict change in aggregation rate upon mutation [12–14]. Further, several structure-based models and empirical equations have been proposed to predict aggregation prone regions and change in aggregation propensity/rate due to mutation [11, 15–21]. Agrawal et al.  and Belli et al.  have reviewed several commonly available aggregation prediction tools and discussed their advantages and shortcomings towards different applications.
In this work, we have collected and analyzed 139 amyloid-fibril forming hexa-peptides from the experiments of Lopez de La Paz and Serrano  and Maurer-Stroh et al. . One hundred and sixty eight hexa-peptide sequences that do not form amyloid-fibrils in experiments conducted by the above mentioned groups were also used. For simplicity, we refer to the two sets as amyloid peptides and non-amyloid peptides. The hexa-peptides in the two datasets are highly homologous as an amyloid peptide may differ from its non-amyloid cousin by just one residue. This indicates that the sequence-structural-thermodynamic features separating amyloid peptides from non-amyloid ones are subtle. Availability of these hexa-peptide sequences with experimental data has afforded us an opportunity to uncover these subtle differences in a systematic manner. The analyses were carried out for several parameters such as overall amino acid composition of the hexa-peptides, preferences for different amino acid residues to occur at each of the six positions in the hexa-peptides and 49 diverse amino acid properties (http://www.cbrc.jp/~gromiha/fold_rate/property.html; [23, 24]). Furthermore, we have developed a set of energy potentials based on the propensity of the 20 amino acid residues at six different positions. We found that amyloid peptides show distinct preferences and avoidances for amino acid residues to occur at each of the six positions. These preferences are significantly different from those seen in non-amyloid peptides. Further, we derived energy potentials based on amino acid preferences at different positions in the hexa-peptides and attempted to use them for discriminating amyloid peptides from non-amyloid ones. The success rate for the energy potentials developed to identify amyloid peptides was 89%. This rate compares favorably with that of position specific matrices based program, Waltz, (67%) , the structure based program, 3Dprofile (80%) , an energy potential based program, PASTA (80%)  and statistical mechanics based program, Tango (91%) . On the other hand, the accuracy for negative prediction, that is, prediction of non-amyloid peptides was only 54%. As far as we know, previous studies have not attempted negative predictions and there is no data to compare this negative prediction rate. Further, we have utilized several machine learning algorithms and the method based on random forest discriminated the amyloid and non-amyloid peptides with an accuracy of 82.7% which is a balance between sensitivity (81.3%) and specificity (83.9%).
Materials and methods
Collection of amyloid peptide and non-amyloid peptide datasets
We have searched the literature as well as the datasets used in previous works to construct a reliable dataset containing peptide sequences that have been studied experimentally for amyloid-fibril formation. For the purpose of this study, we restricted to hexa-peptide sequences verified with experimental data. This procedure yielded a majority of data from Waltz  and Amylhex . In addition, we have used the data reported in the supplementary material of Maurer-Stroh et al. . After eliminating the redundant data, our final dataset contains 139 amyloid forming peptides and 168 non-amyloid peptides.
Amino acid composition
where i stands for the 20 amino acid residues. ni is the number of residues of each type and N is the total number of residues. The summation is through all the residues in all the considered peptides.
where, i and j represent 20 amino acid residues and 6 positions, respectively. N(j) is the total number of residues at position j (i.e. 139 for amyloid and 168 for non-amyloid).
Position specific amino acid propensities
where, Compglob(i) is the composition of residue i obtained with a set of globular proteins [26, 28]. The values are Ala: 8.47; Asp: 5.97; Cys: 1.39; Glu: 6.32; Phe: 3.91; Gly: 7.82; His: 2.26; Ile: 5.71; Lys: 5.76; Leu: 8.48; Met: 2.21; Asn: 4.54; Pro: 4.63; Gln: 3.82; Arg: 4.93; Ser: 5.94; Thr: 5.79; Val: 7.02; Trp: 1.44 and Tyr: 3.58.
where, i and j are the 20 amino acid residues and six positions respectively.
Amino acid properties
In this work, we used a set of 49 diverse amino acid properties (physical, chemical, energetic and conformational). These properties have been used in several studies for understanding protein stability, transition state structures of proteins, and predicting protein folding and unfolding rates, discrimination of transporters and structure-function relationship in membrane proteins [29–35]. The numerical values for all the 49 properties used in this study along with their brief descriptions have been explained in our earlier article [23, 24] and are freely available at http://www.cbrc.jp/~gromiha/fold_rate/property.html. Besides these properties, we also used a hydrophobicity scale based on retention times of individual amino acids in hydrophobic RPLC columns to compute total hydrophobicity values (HT) for each hexa-peptide. This scale is different from all others as it measures latent hydrophobicity of each amino acid . The hydrophobicity coefficient values (h) for each amino acid are as follows: Ala, 13.26; Cys, 26.84; Asp, 8.15; Glu, 11.12; Phe, 90.17; Gly, 3.80; His, 4.14; Ile, 69.53; Lys, 2.92; Leu, 73.84; Met, 51.64; Asn, 1.00; Pro, 27.54; Gln, 6.00; Arg, 10.24; Ser, 3.53; Thr, 11.64; Val, 44.60; Trp, 100.00; Tyr, 47.49; data taken from Table II in . The HT value for each hexa-peptide was calculated by summing the hydrophobicity coefficients of the amino acid residues in the hexa-peptide.
Computation of total amino acid property
where, P(i,j) is the property value of jth residue for the ith peptide and the summation is over 6, the total number of residues in a hexa-peptide. We have repeated the computations for all the 49 amino acid properties in the dataset of amyloid and non-amyloid peptides and the difference between them.
Discrimination of amyloid and non-amyloid peptides using statistically derived energy potentials
If Δϕ (k) has negative value, the peptide is predicted to form amyloid fibrils. Otherwise, it is predicted not to form amyloid-fibrils.
Machine learning techniques for discriminating amyloid and non-amyloid peptides
We have analyzed several machine learning techniques implemented in WEKA program  for discriminating between amyloid and non-amyloid peptides. WEKA includes several methods based on different machine learning techniques such as Bayesian function, Neural network, Radial basis function network, Logistic function, Support vector machine, Regression analysis, Nearest neighbor, Meta learning, Decision tree and Rules. The details of all these methods are available in our earlier articles . We have used the energy potentials and selected amino acid properties as input features for the methods.
Assessment of predictive ability
We have performed 20-fold, 10-fold and 5-fold cross-validation tests for assessing the validity of the present work. In this method, the data set is divided into n groups, n-1 of them are used for training and the rest is used for testing the method. The same procedure is repeated for n times so that each data is used at least once in the test.
We have used different measures, such as sensitivity, specificity and accuracy, to assess the performance of machine learning methods towards discriminating between amyloid and non-amyloid peptides. The term sensitivity shows the correct prediction of amyloid peptides, specificity is the correct prediction of non-amyloid peptides and accuracy indicates the overall assessment. These terms are defined as follows:
Sensitivity = TP/(TP+FN)
Specificity = TN/(TN+FP)
Accuracy = (TP+TN)/(TP+TN+FP+FN),
where, TP (amyloid peptides predicted as amyloid peptides), FP (non-amyloid peptides predicted as amyloid peptides), TN (non-amyloid peptides predicted as non-amyloid peptides) and FN (amyloid peptides predicted as non-amyloid peptides) refer to the number of true positives, false positives, true negatives and false negatives, respectively.
Amino acid composition at different positions of amyloid and non-amyloid peptides
Preferred and avoided residues at different positions of amyloid and non-amyloid forming hexa-peptides
S, W, Y, H, M, N
F, H, K, S, W, Y
C, F, I, Q, T, Y
E, M, T, Y
F, Q, V, W
C, E, F, N, V, W
I, L, N, W, Y
F, I, L, Y
F, I, Y
F, I, Y
C, E, F, M, Q, S, Y
E, F, M, T, W
D, T, R, E
G, L, A, Q, V
D, A, N, S, G
D, N, P, R
A, G, D, I, Y
A, G, R, T, Y
V, S, K, A
A, M, V, D, W
E, L, Q, S, T
A, G, M, R, S, T
A, D, G, I, K
The difference between amyloid and non-amyloid peptides lies in composition of their core positions. The six positions in the hexa-peptides can be divided into two groups consisting termini (positions 1 and 6) and core (positions 2,3,4 and 5). Most of the preferred residues at the core positions in the amyloid peptides are aromatic or aliphatic. The polar amino acids, Asn and Gln are also preferred at positions 2,3 and 4 of amyloid peptides. In contrast, core positions in non-amyloid peptides can contain charged residues also. At the core positions, several avoided residues are common between amyloid and non-amyloid peptides (Table 1). Taken together, the above observations have revealed the differences between amyloid and non-amyloid peptides and suggested that it may be feasible to discriminate between them at the sequence level via computational means. However, the amino acid composition biases seem to suggest that it may be easier to predict that a given hexa-peptide will from amyloid-fibrils than to predict that the given hexa-peptide will not from amyloid fibril (see below). This observation is consistent with the view that amyloid-fibril formation is a backbone-driven process .
Energy potentials derived to amino acid propensities
Energy potentials (kcal/mole) computed from the propensities of amino acid residues at different positions in amyloid and non-amyloid hexa-peptides
Average hydrophobicity of amyloid and non-amyloid forming hexa peptides
Variation of the 49-amino acid properties between amyloid and non-amyloid forming peptides
Average values for the 49 properties used in the present study
Discrimination between amyloid and non-amyloid forming peptides
Can the above described position specific sequence features distinguish amyloid-fibril forming peptides from their close homologues that do not from the amyloid-fibrils? We have made an attempt to discriminate amyloid and non-amyloid forming peptides using hydrophobicity, 49 different properties and the energy potentials. Discrimination based on the energy differentials performed better than the other properties. We devised a statistical method to discriminate amyloid and non-amyloid peptides using total potential computed with Eqn. 6. For each peptide, k, the total energy ϕ(k) was computed for both amyloid-fibril formation and not (from non-amyloid potential). The difference between the energy potentials yield the discriminator, Δϕ (k) (see eqn. 7) The results showed that amyloid peptides are well discriminated with an accuracy of 89% (123/139). This value compares favorably with other prediction methods [10, 20, 25]. However, the discriminator yielded only marginal performance for non-amyloid peptides (accuracy: 54%). Taken together, these results indicate that for a given short peptide sequence, the prediction that it will form amyloid fibrils is easier to make than otherwise. That is, it is harder to predict that the peptide will not form amyloid fibrils. These and other aspects of our work are discussed in the "Discussion" section.
Use of machine learning techniques for discrimination between amyloid and non-amyloid peptides
We have utilized several machine learning techniques for discriminating between amyloid and non-amyloid peptides as described in the Methods section. Overall, most algorithms showed similar performance and the method based on Random forest performed the best. In a 10-fold cross-validation exercise, this method yielded an accuracy of 82.1% when the statistically derived position-specific energy potentials were used. The sensitivity and specificity are 79.9% and 83.9%, respectively. Combining these energy potentials with three amino acid properties, hydrophobicity, isoelectric point and long-range non-bonded energy improved the accuracy marginally to 82.7% (sensitivity, 81.3%; specificity, 83.9%). The method was also tested with 5-fold and 20-fold cross-validations and the accuracies are 80% and 81.1%, respectively.
Fibril forming portions of many amyloidogenic proteins have been traced to short peptides by many experimental groups. The smallest length for a peptide that forms amyloid fibrils is three amino acid residues. Tetra-peptides have also been shown to form amyloid-fibrils . The most common sequence lengths for the amyloid-fibril peptides are 5-9 residues. We chose to focus on hexa-peptides because hexa-peptides have been often used in experiments to grow amyloid-fibrils [10, 11]. Available experimental data on short peptides that form amyloid fibrils and those that do not form amyloid fibrils shows that even single residue differences are important [10, 11]. With the growing interest in nano-materials made out of peptide aggregates with well-defined fibrillar morphologies and desirable properties , it has become important to computationally predict which of the short peptide sequences are capable of forming amyloid fibrils with desired properties and which of them would not yield such fibrils, even though they may form amorphous β-aggregates and may still contain kernels of the cross-β steric zipper motif. To our knowledge, this is the first attempt to discriminate between amyloid fibril and non-amyloid fibril forming peptide sequences using empirical/computational means. Here, we have used publicly available information on Amyl Hex and other hexa-peptides to uncover the subtle sequence-structural features that could be different between amyloid and non-amyloid peptides. The sequences in the two peptide datasets are highly homologous and almost all non-amyloid peptides do form amorphous β-aggregates . Thus, it was not surprising that almost all of the 49 physico-chemical amino acid properties  showed only small differences between amyloid and non-amyloid peptide sets. Despite the high sequence homologies, the overall and position-wise amino acid propensities are different between amyloid and non-amyloid peptide sets. This indicates that amino acid side chains do play a role in amyloid-fibril formation even though the process has been thought to be mainly driven by backbone . The differences in sequence features and positional context in formation of amyloid and non-amyloid peptides were converted in to the energy potentials in this study. These potentials were able to successfully identify the amyloid-peptides in most cases. This validates our approach for predicting potential amyloidogenic sequences.
Almost all studies in computational biology focus on making positive predictions. However, in this study we attempted to make negative predictions also. That is, we tried to predict that a given peptide will not from amyloid fibrils, even though it may self-associate via cross-β motif and form amorphous β-aggregates. In this case, making the negative prediction proved to be harder than the positive prediction that a peptide will form amyloid-fibril. There could be several reasons for this. First of all, kinetics of amyloid fibril formation depend on critical monomer concentrations required to initiate the process . The critical monomer concentrations required for initiation of fibril formation were found to vary in the range of 30-400 μM for highly homologous tetra-peptides, KFFE, KVVE, KLLE [12, 44]. However, the experiments that determine which peptides in a given set form amyloid fibrils use a single concentration value for all the peptides. The time periods over which fibril are grown are also arbitrarily set (see method section in [10–12]). These experimental condition requirements imply that the peptides which require higher critical monomer concentrations to initiate fibril formation and/or which have slower fibril growth kinetics may be falsely designated as non-amyloid peptides even-though their sequences may contain all the required features for amyloid-fibril formation. Secondly, the preferences for the individual amino acid residues to occur at each of the six positions are better characterized for amyloid and non-amyloid peptides than avoidances (Table 2), that is, amyloid fibril forming sequence features are based on positive, and not negative, selection of the relevant physicochemical properties at the level of individual amino acids. Third, the peptide sequences in the two data set are highly homologous and most of the peptides in the non-amyloid set, derived from the parent amyloid-fibril forming peptide sequence, STVIIE  form β-aggregates. It is quite probable that many of the peptides in non-amyloid set would form amyloid-fibrils in slightly different experimental conditions.
The prediction for non-amyloid peptides improved when three amino acid properties, namely, hydrophobicity, isoelectric point and long range interaction energy, were combined with the position specific energy potentials and machine learning techniques were used. In such techniques, the data are trained so that the methods should perform equally well for both amyloids and non-amyloid peptide sets. Not surprisingly, this procedure showed similar levels of sensitivity and specificity, and the accuracy is the balance between these two terms.
We have analyzed the available experimental data on the hexa-peptides that from amyloid fibrils and on those that do not form amyloid-fibrils. We found that amyloid peptides show position-specific preference and avoidances that are different from their homologues which may form β-aggregates but not fibrils. These position-specific preferences of amino acid residues have been utilized to discriminate amyloid forming peptides and non-amyloids using statistical methods and machine learning techniques. In the next step, we plan to combine single residue propensities with the residue pair propensities in a position wise manner to further improve our ability to predict both amyloid and non-amyloid forming hexa-peptides.
AMT and DV thank the Bioinformatics Facility of University of Madras for computational facilities. SK acknowledges his discussions with Drs. Patrick Buck and Satish Singh on topics related to aggregation in proteins and peptides. MMG wishes to thank IIT Madras for infrastructure facilities. The work was partially supported by the Department of Biotechnology, Government of India (BT/PR7150/BID/7/424/2012).
The publication charges are funded by the Indian Institute of Technology Madras.
This article has been published as part of BMC Bioinformatics Volume 14 Supplement 8, 2013: Proceedings of the 2012 International Conference on Intelligent Computing (ICIC 2012). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcbioinformatics/supplements/14/S8.
- Agrawal NJ, Kumar S, Wang X, Helk B, Singh SK, Trout BL: Aggregation in protein-based biotherapeutics: Computational studies and tools to identify aggregation-prone regions. J Pharm Sci. 2011, 100: 5081-5095. 10.1002/jps.22705.View ArticlePubMedGoogle Scholar
- Webb RL, Murphy MP: β-Secretases, Alzheimer's Disease, and Down Syndrome. 2012, Curr Gerontol Geriatr Res 2012, 362839-Google Scholar
- Checler F, Turner AJ: Journal of Neurochemistry special issue on Alzheimer's disease: 'amyloid cascade hypothesis--20 years on. J Neurochem. 2012, 120 (Suppl 1): iii-iv.View ArticlePubMedGoogle Scholar
- Wiltzius JJ, Landau M, Nelson R, Sawaya MR, Apostol MI, Goldschmidt L, Soriaga AB, Cascio D, Rajashankar K, Eisenberg D: Molecular mechanisms for protein-encoded inheritance. Nat Struct Mol Biol. 2009, 16: 973-978. 10.1038/nsmb.1643.PubMed CentralView ArticlePubMedGoogle Scholar
- Colletier JP, Laganowsky A, Landau M, Zhao M, Soriaga AB, Goldschmidt L, Flot D, Cascio D, Sawaya MR, Eisenberg D: Molecular basis for amyloid-beta polymorphism. Proc Natl Acad Sci USA. 2011, 108: 16938-43. 10.1073/pnas.1112600108.PubMed CentralView ArticlePubMedGoogle Scholar
- Berman HM, Kleywegt GJ, Nakamura H, Markley JL: The Protein Data Bank at 40: reflecting on the past to prepare for the future. Structure. 2012, 20: 391-6. 10.1016/j.str.2012.01.010.PubMed CentralView ArticlePubMedGoogle Scholar
- Chiti F, Taddei N, Baroni F, Capanni C, Stefani M, Ramponi G: Kinetic partitioning of protein folding and aggregation. Nat Struct Mol Biol. 2002, 9: 137-43. 10.1038/nsb752.View ArticleGoogle Scholar
- Nelson R, Sawaya MR, Balbirnie M, Madsen AO, Riekel C, Grothe R, Eisenberg D: Structure of the cross-beta spine of amyloid-like fibrils. Nature. 2005, 435: 773-8. 10.1038/nature03680.PubMed CentralView ArticlePubMedGoogle Scholar
- Sawaya MR, Sambashivan S, Nelson R, Ivanova MI, Sievers SA, Apostol MI, Thompson MJ, Balbirnie M, Wiltzius JJ, McFarlane HT, Madsen AO, Riekel C, Eisenberg D: Atomic structures of amyloid cross-beta spines reveal varied steric zippers. Nature. 2007, 447: 453-7. 10.1038/nature05695.View ArticlePubMedGoogle Scholar
- Lopez de la Paz M, Serrano L: Sequence determinants of amyloid fibril formation. Proc Natl Acad Sci USA. 2004, 101: 87-92. 10.1073/pnas.2634884100.PubMed CentralView ArticlePubMedGoogle Scholar
- Maurer-Stroh S, Debulpaep M, Kuemmerer N, Lopez de la Paz M, Martins IC, Reumers J, Morris KL, Copland A, Serpell L, Serrano L, Schymkowitz JW, Rousseau F: Exploring the sequence determinants of amyloid structure using position-specific scoring matrices. Nat Methods. 2010, 7: 237-42. 10.1038/nmeth.1432.View ArticlePubMedGoogle Scholar
- Tjernberg L, Hosia W, Bark N, Thyberg J, Johansson J: Charge attraction and beta propensity are necessary for amyloid fibril formation from tetrapeptides. J Biol Chem. 2002, 277: 43243-6. 10.1074/jbc.M205570200.View ArticlePubMedGoogle Scholar
- Tartaglia GG, Cavalli A, Pellarin R, Caflisch A: The role of aromaticity, exposed surface, and dipole moment in determining protein aggregation rates. Protein Sci. 2004, 13: 1939-1941. 10.1110/ps.04663504.PubMed CentralView ArticlePubMedGoogle Scholar
- de Groot NS, Aviles FX, Vendrell J, Ventura S: Mutagenesis of the central hydrophobic cluster in Aβ42 Alzheimer's peptide. Side-chain properties correlate with aggregation propensities. FEBS J. 2006, 273: 658-668. 10.1111/j.1742-4658.2005.05102.x.View ArticlePubMedGoogle Scholar
- DuBay KF, Pawar AP, Chiti F, Zurdo J, Dobson CM, Vendruscolo M: Prediction of the absolute aggregation rates of amyloidogenic polypeptide chains. J Mol Biol. 2004, 341: 1317-1326. 10.1016/j.jmb.2004.06.043.View ArticlePubMedGoogle Scholar
- Fernandez-Escamilla AM, Rousseau F, Schymkowitz J, Serrano L: Prediction of sequence-dependent and mutational effects on the aggregation of peptides and proteins. Nat Biotechnol. 2004, 22: 1302-1306. 10.1038/nbt1012.View ArticlePubMedGoogle Scholar
- Pawar AP, Dubay KF, Zurdo J, Chiti F, Vendruscolo M, Dobson CM: Prediction of "aggregation-prone" and "aggregation-susceptible" regions in proteins associated with neurodegenerative diseases. J Mol Biol. 2005, 350: 379-392. 10.1016/j.jmb.2005.04.016.View ArticlePubMedGoogle Scholar
- Tartaglia GG, Cavalli A, Pellarin R, Caflisch A: Prediction of aggregation rate and aggregation-prone segments in polypeptide sequences. Protein Science. 2005, 14: 2723-2734. 10.1110/ps.051471205.PubMed CentralView ArticlePubMedGoogle Scholar
- Galzitskaya OV, Garbuzynskiy SO, Lobanov MY: Prediction of amyloidogenic and disordered regions in protein chains. PLoS Comput Biol. 2006, 2: e177-10.1371/journal.pcbi.0020177.PubMed CentralView ArticlePubMedGoogle Scholar
- Thompson MJ, Sievers SA, Karanicolas J, Ivanova MI, Baker D, Eisenberg D: The 3D profile method for identifying fibril-forming segments of proteins. Proc Natl Acad Sci USA. 2006, 103: 4074-4078. 10.1073/pnas.0511295103.PubMed CentralView ArticlePubMedGoogle Scholar
- Zibaee S, Makin OS, Goedert M, Serpell LC: A simple algorithm locates β-strands in the amyloid fibril core of α-synuclein, Aβ, and tau using the amino acid sequence alone. Protein Sci. 2007, 16: 906-918. 10.1110/ps.062624507.PubMed CentralView ArticlePubMedGoogle Scholar
- Belli M, Ramazzotti M, Chiti F: Prediction of amyloid aggregation in vivo. EMBO Rep. 2011, 12: 657-63. 10.1038/embor.2011.116.PubMed CentralView ArticlePubMedGoogle Scholar
- Gromiha MM, Oobatake M, Sarai A: Important amino acid properties for enhanced thermostability from mesophilic to thermophilic proteins. Biophys Chem. 1999, 82: 51-67. 10.1016/S0301-4622(99)00103-9.View ArticlePubMedGoogle Scholar
- Gromiha MM: A statistical model for predicting protein folding rates from amino acid sequence with structural class information. J Chem Inf Model. 2005, 45: 494-501. 10.1021/ci049757q.View ArticlePubMedGoogle Scholar
- Trovato A, Seno F, Tosatto SCE: The PASTA server for protein aggregation prediction. Protein Eng Des Sel. 2007, 20: 521-523. 10.1093/protein/gzm042.View ArticlePubMedGoogle Scholar
- Gromiha MM, Suwa M: A simple statistical method for discriminating outer membrane proteins with better accuracy. Bioinformatics. 2005, 21: 961-8. 10.1093/bioinformatics/bti126.View ArticlePubMedGoogle Scholar
- Chou PY, Fasman GD: Conformational parameters for amino acids in helical, beta-sheet, and random coil regions calculated from proteins. Biochemistry. 1974, 13: 211-22. 10.1021/bi00699a001.View ArticlePubMedGoogle Scholar
- Gromiha MM: Protein Bioinformatics: From Sequence to Function. 2010, Elsevier/Academic PressGoogle Scholar
- Gromiha MM, Oobatake M, Kono H, Uedaira H, Sarai A: Role of structural and sequence information in the prediction of protein stability changes: comparison between buried and partially buried mutations. Protein Eng. 1999, 12: 549-555. 10.1093/protein/12.7.549.View ArticlePubMedGoogle Scholar
- Gromiha MM, Selvaraj S: Important amino acid properties for determining the transition state structures of two-state protein mutants. FEBS Lett. 2002, 526: 129-134. 10.1016/S0014-5793(02)03122-8.View ArticlePubMedGoogle Scholar
- Gromiha MM, Selvaraj S: Inter-residue interactions in protein folding and stability. Prog Biophys Mol Biol. 2004, 86: 235-77. 10.1016/j.pbiomolbio.2003.09.003.View ArticlePubMedGoogle Scholar
- Gromiha MM, Thangakani AM, Selvaraj S: FOLD-RATE: Prediction of Protein Folding Rates from Amino Acid Sequence. Nucleic Acids Res. 2006, 34: W70-W74. 10.1093/nar/gkl043.PubMed CentralView ArticlePubMedGoogle Scholar
- Gromiha MM, Selvaraj S, Thangakani AM: A statistical method for predicting protein unfolding rates from amino acid sequence. J Chem Inf Model. 2006, 46: 1503-1508. 10.1021/ci050417u.View ArticlePubMedGoogle Scholar
- Chen SA, Ou YY, Lee TY, Gromiha MM: Prediction of transporter targets using efficient RBF networks with PSSM profiles and biochemical properties. Bioinformatics. 2011, 27: 2062-7. 10.1093/bioinformatics/btr340.View ArticlePubMedGoogle Scholar
- Gromiha MM, Harini K, Sowdhamini R, Fukui K: Relationship between amino acid properties and functional parameters in olfactory receptors and discrimination of mutants with enhanced specificity. BMC Bioinformatics. 2012, 13 (Suppl 7): S1-10.1186/1471-2105-13-S7-S1.PubMed CentralView ArticlePubMedGoogle Scholar
- Mant CT, Kovacs JM, Kim HM, Pollock DD, Hodges RS: Intrinsic amino acid side-chain hydrophilicity/hydrophobicity coefficients determined by reversed-phase high-performance liquid chromatography of model peptides: comparison with other hydrophilicity/hydrophobicity scales. Biopolymers. 2009, 92: 573-95. 10.1002/bip.21316.PubMed CentralView ArticlePubMedGoogle Scholar
- Gromiha MM, Suwa M: Influence of amino acid properties for discriminating outer membrane proteins at better accuracy. Biochim Biophys Acta. 2006, 1764: 1493-7. 10.1016/j.bbapap.2006.07.005.View ArticlePubMedGoogle Scholar
- Witten IH, Frank E: Data Mining: Practical machine learning tools and techniques. 2005, Morgan Kaufmann, San Francisco, 2Google Scholar
- Kim W, Hecht MH: Generic hydrophobic residues are sufficient to promote aggregation of the Alzheimer's Abeta42 peptide. Proc Natl Acad Sci USA. 2006, 103: 15824-9. 10.1073/pnas.0605629103.PubMed CentralView ArticlePubMedGoogle Scholar
- Luheshi LM, Tartaglia GG, Brorsson AC, Pawar AP, Watson IE, Chiti F, Vendruscolo M, Lomas DA, Dobson CM, Crowther DC: Systematic in vivo analysis of the intrinsic determinants of amyloid Beta pathogenicity. PLoS Biol. 2007, 5: e290-10.1371/journal.pbio.0050290.PubMed CentralView ArticlePubMedGoogle Scholar
- Eakin CM, Berman AJ, Miranker AD: A native to amyloidogenic transition regulated by a backbone trigger. Nature Str Mol Biol. 2006, 13: 202-208. 10.1038/nsmb1068.View ArticleGoogle Scholar
- Gromiha MM, Thangakani AM, Kumar S, Velmurugan D: Sequence Analysis and Discrimination of Amyloid and Non-amyloid Peptides. Comm Comp Inf Sci. 304: 447-452.Google Scholar
- Naskar J, Drew MG, Deb I, Das S, Banerjee A: Water-soluble tripeptide Abeta (9-11) forms amyloid-like fibrils and exhibits neurotoxicity. Org Lett. 2008, 10: 2625-8. 10.1021/ol8007217.View ArticlePubMedGoogle Scholar
- Knowles TPJ, Buehler MJ: Nanomechanics of functional and pathological amyloid materials. Nature Nanotech. 2011, 6: 469-479. 10.1038/nnano.2011.102.View ArticleGoogle Scholar
- Wolf MG, Jongejan JA, Laman JD, de Leeuw SW: Quantitative prediction of amyloid fibril growth of short peptides from simulations: Calculating association constants to dissect side chain importance. J Amer Chem Soc. 2008, 130: 15772-15773. 10.1021/ja806606y.View ArticleGoogle Scholar
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.