Genes contain instructions to make proteins from building blocks called amino acids. The instructions are encoded in units called codons that each specify a single amino acid in the chain. A small mutation in a particular codon can change the amino acid found at the corresponding position in the protein. Some amino acids interact with other amino acids in the chain, thereby enabling the protein to adopt the three-dimensional shape it needs to work properly. Therefore, a mutation that affects one of these amino acids may have a large impact on the ability of the protein to work.

A mutation at one position in the protein may, however, have little effect if it is accompanied by a ‘compensatory’ mutation at another position. Such compensatory mutations are more likely to occur when the two positions in the protein are close to each other. To identify such mutations, the amino acid sequences of similar proteins from different organisms are aligned and compared.

A computational method called ‘correlated mutation analysis’ searches for pairs of positions in the alignment that display co-variation, i.e. where particular mutations at one position tend to be accompanied by certain mutations at the second position. These pairs are then ranked according to the strength of their correlation and those with the highest ranking are predicted to be in close contact. Such predictions are, however, far from perfect and can give false results.

Jacob et al. developed and tested a new technique of correlated mutation analysis by examining codon sequences as well as amino acid sequences. The rationale behind the technique relies on the fact that several different codons can encode the same amino acid, so that a mutation in a codon does not always change the amino acid it encodes. Therefore, a strong correlation at the amino acid level can be accompanied by a weak correlation at the codon level. In such cases the positions are more likely to be in contact than in cases where there is a strong correlation also at the codon level since the correlation can then be due to constraints at the DNA or RNA level.

Jacob et al. tested their approach using different methods for analyzing correlated mutations that were proposed in previous studies. This showed that the predictions obtained using both amino acid and codon data are significantly more accurate than those obtained by comparing amino acid sequences only. Future work will test whether combining amino acid and codon data can also be used to predict interactions between different proteins.

The effects of mutations that disrupt protein structure and/or function at one site are often suppressed by mutations that occur at other sites either in the same protein or in other proteins. Such compensatory mutations can occur at positions that are distant from each other in space, thus, reflecting long-range interactions in proteins (Horovitz et al., 1994; Lee et al., 2008). It has often been assumed, however, that most compensatory mutations occur at positions that are close in space, thus motivating the development of computational methods for identifying co-evolving positions as distance constraints in protein structure prediction (Göbel et al., 1994). These methods, which rely on multiple sequence alignments (MSA) of homologous proteins as inputs, will become increasingly more useful in the coming years owing to the explosive growth in sequence data. The output of methods for correlated mutation analysis (CMA) is a rank order of the pairs of columns in the alignment according to the statistical and/or physical signficance attached to the correlation observed for each pair. The various methods for CMA that have been developed in the past 15 years differ in the measures they employ for attaching significance to the correlations (Livesay et al., 2012; de Juan et al., 2013; Mao et al., 2015). Early measures include, for example, mutual information (MI) from information theory (Gloor et al., 2005) and observed-minus-expected-squared (OMES) in the chi-square test (Kass and Horovitz, 2002).

Statistically significant correlations in MSAs that do not reflect interactions between residues in contact, that is, false positives, can stem from (i) various indirect physical interactions and (ii) common ancestry. The extent of false positives due to the latter source is manifested in the large number of correlations between positions in non-interacting proteins that can be observed when the sequences of non-interacting proteins from the same organism are concatanated and subjected to CMA (Noivirt et al., 2005). Several methods for removing false positives owing to common ancestry were developed (Pollock et al., 1999; Wollenberg and Atchley, 2000; Noivirt et al., 2005; Dunn et al., 2008) but their success in contact prediction using CMA remained limited. False positives due to the former source, that is, indirect physical interactions, can occur when, for example, correlations corrersponding to positions i and j that are in contact and positions j and k that are in contact lead to a correlation for positions i and k that are not in contact. Methods that remove such transitive correlations have been developed in recent years and include, for example, Direct Coupling Analysis (DCA or DI for Direct Information) (Weigt et al., 2009; Morcos et al., 2011), Protein Sparse Inverse COVariance (PSICOV) (Jones et al., 2012) and Gremlin's pseudolikelihood method (Kamisetty et al., 2013). These methods have been found to be very successful in identifying contacting residues (Marks et al., 2012) and they outperform earlier methods (Mao et al., 2015). Nevertheless, their accuracy, which is ∼80% for the correlations in the top 0.1% (ranked by their scores), drops to ∼50% for the top 1% (Mao et al., 2015). Given that the number of contacts in a protein with N residues is ∼N (Faure et al., 2008), it follows that for proteins with, for example, 100 residues (i.e. with 4560 potential contacts between residues separated by at least 5 residues in the sequence) only about 25% of the contacts (i.e. 23 of the top 1% 46 predictions) will be identified by these CMA methods. In addition, these methods require large MSAs comprising thousands of sequences in order to perform well and such sequence data are not always available. Consequently, it is clear that much can be gained from further improvements in methods of CMA. Here, we describe the development and application of a new method for CMA that uses both amino acid and codon MSAs as inputs instead of relying exclusively on amino acid MSAs as done before. We show that contact prediction is improved in a meaningful manner when amino acid and codon information are combined.

The key premise underlying the method introduced here is that a correlation at the amino acid level between two positions is more likely to reflect a direct interaction if the correlation at the codon level for these positions is weak (Figure 1). In other words, it is assumed that cases of strong correlations at both the amino acid and codon levels for a pair of positions are less likely to reflect selection to conserve protein contacts and more likely to reflect selection to conserve interactions involving DNA or RNA and/or common ancestry. Given this rationale in mind, we decided to test whether contact identification improves when all the pairs of positions are ranked using a score that increases with (i) increasing strength of the correlation at the amino acid level and (ii) decreasing strength of the correlation at the codon level. Such a score, S_i, is given, for example, by: S_i = S_i^α(AA)/S_i(C), where S_i(AA) and S_i(C) are the scores generated by method i (e.g., MI) for the amino acid and codon alignments, respectively, and the value of the power α is determined empirically depending on the method (see below).

Figure 1

Download asset Open asset

The approach outlined above was tested for the OMES (Kass and Horovitz, 2002), MI (Gloor et al., 2005), MIp (Dunn et al., 2008) and DCA (Morcos et al., 2011) methods using 114 MSAs each comprising at least 2000 sequences of length between 200 and 500 residues. In the case of the PSICOV method (Jones et al., 2012), only 86 MSAs out of the 114 MSAs were used since the others didn't pass this method's threshold for amino acid sequence diversity. Each MSA also included at least one sequence with a known crystal structure at a resolution <3 Å in which at least 80% of all the residues are resolved. The mean accuracy of contact identification was plotted as a function of the top ranked number of predicted pairwise contacts (Figure 2—figure supplement 1) or as a function of the top ranked fraction of protein length, L, number of predicted pairwise contacts (Figure 2). Residues were considered as being in contact if the distance between their C_β atoms is ≤8 Å following the definition used in CASP experiments (Ezkurdia et al., 2009) and other studies (Kamisetty et al., 2013; Skwark et al., 2014) (see also Figure 2—figure supplement 2). The results show that the PSICOV and DCA methods outperform the OMES, MI and MIp methods (Figure 2) as established before (Mao et al., 2015). They also show that combining amino acid and codon data leads to an improvement in the predictions by OMES, MI, DCA and PSICOV. In the case of MIp, however, no improvement is observed despite the fact that this method performs worse than DCA and PSICOV. In MIp, a term called average product correction (APC) is subtracted from the MI score for each pair of positions in order to reduce false positives. Removing this correction from PSICOV where it also exists and including the codon data yielded the best method (Figure 2). Hence, we can conclude that there is an overlap between the background noise reduced upon including the APC term and codon data and that including the latter can be more advantageous as we observe for PSICOV.

Figure 2 with 2 supplements see all

Download asset Open asset

The extent of improvement increases with increasing values of the power α until a maximum is reached (Figure 3A) at a value of α that depends on the method used and different values of α_max were, therefore, chosen accordingly. Cross-validation by dividing the MSA data into training and test sets showed that the values of α_max are stable, that is, they do not vary depending on the set of MSAs (Figure 3—figure supplement 1). Given these values of α _max, the significance of the extent of improvement was assessed by comparing for each MSA the accuracy of the contact predictions using the different methods with and without incorporating codon data. Significance levels were determined using two non-parametric tests: (i) the Wilcoxon signed-rank test, which takes into account both the number of MSAs for which the accuracy of the contact predictions increases upon incorporating codon data (e.g., 81 in the case of DCA) and the magnitude of the improvement; and (ii) the sign test, which only considers the number of MSAs with improved accuracy. The extent of improvement achieved by incorporating codon data was found to be highly significant as indicated by the p-values obtained using both tests (Figure 3B).

Figure 3 with 1 supplement see all

Download asset Open asset

The improvement in the predictions upon combining amino acid and codon data, when residues are defined as being in contact if the distance between their C_β atoms is ≤8 Å, led us to examine whether direct contacts are identified better using this contact definition compared with an ‘All’ definition used by others (Morcos et al., 2011) according to which a contact exists if at least one inter-atomic distance between the residues is ≤8 Å. A non-redundant set of 2481 proteins with an available crystal structure at a resolution better than 1.6 Å was compiled and all the residue pairs in each structure that are in contact according to these two definitions were identified. We then determined for each protein what is the fraction of the residue pairs in contact according to these two definitions that are actually in direct physical contact, that is, with a distance <3.5 Å between two of their respective heavy atoms. It should be noted that atoms can interact with each other even if the distance between them is larger than 3.5 Å via, for example, weak electrostatic interactions but pairs of atoms which are closer than 3.5 Å can always be considered as interacting. It may be seen that, on average, pairs in direct contact constitute only about 10% of the pairs in contact according to the ‘All’ definition and 30% of the pairs in contact according to the C_β-based definition (Figure 2—figure supplement 2). The better success of DCA in identifying contacts according to the C_β-based definition when amino acid and codon data are combined is, therefore, an important result since more pairs that are in true physical contact are identified in this way.

Our finding that the C_β-based definition of contacts is better than the ‘All’ definition but still poor (only 30% of the pairs defined as being in contact are in physical contact) prompted us to test the performance of our method for additional contact definitions. The mean of the extent of improvement in contact prediction for 114 domains (or 86 in the case of PSICOV) was, therefore, determined as a function of the distance that must exist between at least two C_β atoms in different residues in order for them to be defined as being in contact. It may be seen that, in the cases of PSICOV, OMES and DCA, the maximum improvements in contact prediction upon combining amino acid and codon data are when these distances are about 5.5, 7 and 5.5 Å, respectively, and that, in the cases of DCA and OMES, the improvement decreases dramatically when this distance is >∼10 Å (Figure 4). In the case of MI, the extent of improvement upon combining amino acid and codon data is found to be relatively insensitive to the distance used to define a contact and is maximal when it is ∼4.5 Å (Figure 4). These data, therefore, show again that the improvement in contact prediction upon combining amino acid and codon data is greatest when the distance used for contact definition does not lead to many pairs being defined in contact when in fact they are not in direct physical contact.

Figure 4

Download asset Open asset

The added value in combining amino acid and codon data can be illustrated for contact prediction by DCA in the case of Kex1Δp, a prohormone-processing carboxypeptidase from Saccharomyces cerevisiae that lacks the acidic domain and membrane-spanning portion of Kex1p. The crystal structure of Kex1Δp was solved at a resolution of 2.4 Å (Shilton et al., 1997) and its MSA consists of 1877 sequences. The predictions by DCA with or without incorporating codon data are shown in the respective top and bottom halves of the Kex1Δp contact map (Figure 5A). A comparison of the predictions by the two approaches shows that those made with incorporation of codon data are more long-range (in sequence) and more spread throughout the protein structure than those made without incorporation of codon data. Examples for such long-range contacts between different secondary structure elements in Kex1Δp that are predicted only when also the codon data is used include the interactions between Thr148 with Phe185, Ala186 with Leu208 and Leu190 with Leu368 (Figure 5B). This and other examples (Figure 5—figure supplement 1) show that incorporation of codon data can yield predictions of contacts between residues that are distant in sequence and are, thus, of more value for structure prediction.

Figure 5 with 1 supplement see all

Download asset Open asset

1. 1. Finn RD
   2. Bateman A
   3. Clements J
   4. Coggill P
   5. Eberhardt RY
   6. Eddy SR
   7. Heger A
   8. Hetherington K
   9. Holm L
   10. Mistry J
   11. Sonnhammer EL
   12. Tate J
   13. Punta M

   (2013) Pfam version 27.0

   All domain families at RP75 redundancy level are publicly available at Pfam EMBL-EBI ftp site.

   http://pfam.xfam.org/
2. 1. Pruitt KD
   2. Brown GR
   3. Hiatt SM
   4. Thibaud-Nissen F
   5. Astashyn A
   6. Ermolaeva O
   7. Farrell CM
   8. Hart J
   9. Landrum MJ
   10. McGarvey KM
   11. Murphy MR
   12. O'Leary NA
   13. Pujar S
   14. Rajput B
   15. Rangwala SH
   16. Riddick LD
   17. Shkeda A
   18. Sun H
   19. Tamez P
   20. Tully RE
   21. Wallin C
   22. Webb D
   23. Weber J
   24. Wu W
   25. Dicuccio M
   26. Kitts P
   27. Maglott DR
   28. Murphy TD
   29. Ostell JM

   (2015) RefSeq

   All RefSeq sequences are publicly available at NCBI Reference sequence database using ftp service or EMBL-EBI's WSDbfetch services.

   http://www.ncbi.nlm.nih.gov/refseq/
3. 1. Cunningham F
   2. Ridwan Amode M
   3. Barrell D
   4. Beal K
   5. Billis K
   6. Brent S
   7. Carvalho-Silva D
   8. Clapham P
   9. Coates G
   10. Fitzgerald S
   11. Gil L
   12. Girón CG
   13. Gordon L
   14. Hourlier T
   15. Hunt SE
   16. Janacek SH
   17. Johnson N
   18. Juettemann T
   19. Kähäri AK
   20. Keenan S
   21. Martin FJ
   22. Maurel T
   23. McLaren W
   24. Murphy DN
   25. Nag R
   26. Overduin B
   27. Parker A
   28. Patricio M
   29. Perry E
   30. Pignatelli M
   31. Riat HS
   32. Sheppard D
   33. Taylor K
   34. Thormann A
   35. Vullo A
   36. Wilder SP
   37. Zadissa A
   38. Aken BL
   39. Birney E
   40. Harrow J
   41. Kinsella R
   42. Muffato M
   43. Ruffier M
   44. Searle SMJ
   45. Spudich G
   46. Trevanion SJ
   47. Yates A
   48. Zerbino DR
   49. Flicek P

   (2015) Ensembl

   All Ensemblgenomes sequences are publicly available at ENSEMBLGENOMES site using ftp or REST API.

   http://www.ensembl.org/
4. 1. Berman HM
   2. Westbrook J
   3. Feng Z
   4. Gilliland G
   5. Bhat TN
   6. Weissig H
   7. Shindyalov IN
   8. Bourne PE

   (2015) Protein Data Bank

   All pdbcodes in this work are publicly available at RCSB Protein Data Bank using ftp.

   http://www.rcsb.org/
5. 1. Wang G
   2. Dunbrack RL

   (2015) CullPDB

   All datasets are publicly available at PISCES server home page.

   http://dunbrack.fccc.edu/Guoli/culledpdb/
6. 1. Velankar S
   2. Dana JM
   3. Jacobsen J
   4. van Ginkel G
   5. Gane PJ
   6. Luo J
   7. Oldfield TJ
   8. O'Donovan C
   9. Martin MJ
   10. Kleywegt GJ

   (2015) SIFTS

   All datasets are publicly available at ENSEMBL site.

   ftp://ftp.ebi.ac.uk/pub/databases/msd/sifts/text/
7. 1. The UniProt Consortium

   (2014) Uniprot

   All uniprot entriess are publicly available at uniprot site.

   http://www.uniprot.org/
8. 1. Leinonen R
   2. Akhtar R
   3. Birney E
   4. Bower L
   5. Cerdeno-Tárraga A
   6. Cheng Y
   7. Cleland I
   8. Faruque N
   9. Goodgame N
   10. Gibson R
   11. Hoad G
   12. Jang M
   13. Pakseresht N
   14. Plaister S
   15. Radhakrishnan R
   16. Reddy K
   17. Sobhany S
   18. Ten Hoopen P
   19. Vaughan R
   20. Zalunin V
   21. Cochrane G

   (2015) European Nucleotide Archive

   All sequences are publicly available at EBI ENA site using ftp service or EMBL-EBI's WSDbfetch services.

   http://www.ebi.ac.uk/ena
9. 1. Kersey PJ
   2. Allen JE
   3. Christensen M
   4. Davis P
   5. Falin LJ
   6. Grabmueller C
   7. Hughes DS
   8. Humphrey J
   9. Kerhornou A
   10. Khobova J
   11. Langridge N
   12. McDowall MD
   13. Maheswari U
   14. Maslen G
   15. Nuhn M
   16. Ong CK
   17. Paulini M
   18. Pedro H
   19. Toneva I
   20. Tuli MA
   21. Walts B
   22. Williams G
   23. Wilson D
   24. Youens-Clark K
   25. Monaco MK
   26. Stein J
   27. Wei X
   28. Ware D
   29. Bolser DM
   30. Howe KL
   31. Kulesha E
   32. Lawson D
   33. Staines DM

   (2015) Ensemblegeneomes

   All Ensemblgenomes sequences are publicly available at ENSEMBLGENOMES site using ftp or REST API.

   http://ensemblgenomes.org/

1. 1. Berman HM
   2. Westbrook J
   3. Feng Z
   4. Gilliland G
   5. Bhat TN
   6. Weissig H
   7. Shindyalov IN
   8. Bourne PE

   (2015) Protein Data Bank

   Protein Data Bank, RCSB Protein Data Bank, http://www.rcsb.org/.

   http://www.rcsb.org/

   • Google Scholar
2. 1. Chen C
   2. Natale DA
   3. Finn RD
   4. Huang H
   5. Zhang J
   6. Wu CH
   7. Mazumder R

   (2011) Representative proteomes: a stable scalable and unbiased proteome set for sequence analysis and functional annotation

   PLOS ONE 6:e18910.

   https://doi.org/10.1371/journal.pone.0018910

   • Google Scholar
3. 1. Cunningham F
   2. Ridwan Amode M
   3. Barrell D
   4. Beal K
   5. Billis K
   6. Brent S
   7. Carvalho-Silva D
   8. Clapham P
   9. Coates G
   10. Fitzgerald S
   11. Gil L
   12. Girón CG
   13. Gordon L
   14. Hourlier T
   15. Hunt SE
   16. Janacek SH
   17. Johnson N
   18. Juettemann T
   19. Kähäri AK
   20. Keenan S
   21. Martin FJ
   22. Maurel T
   23. McLaren W
   24. Murphy DN
   25. Nag R
   26. Overduin B
   27. Parker A
   28. Patricio M
   29. Perry E
   30. Pignatelli M
   31. Riat HS
   32. Sheppard D
   33. Taylor K
   34. Thormann A
   35. Vullo A
   36. Wilder SP
   37. Zadissa A
   38. Aken BL
   39. Birney E
   40. Harrow J
   41. Kinsella R
   42. Muffato M
   43. Ruffier M
   44. Searle SMJ
   45. Spudich G
   46. Trevanion SJ
   47. Yates A
   48. Zerbino DR
   49. Flicek P

   (2015) Ensembl

   Ensembl, http://www.ensembl.org/.

   http://www.ensembl.org/

   • Google Scholar
4. 1. Dunn SD
   2. Wahl LM
   3. Gloor GB

   (2008) Mutual information without the influence of phylogeny or entropy dramatically improves residue contact prediction

   Bioinformatics 24:333–340.

   https://doi.org/10.1093/bioinformatics/btm604

   • Google Scholar
5. 1. de Juan D
   2. Pazos F
   3. Valencia A

   (2013) Emerging methods in protein co-evolution

   Nature Reviews Genetics 14:249–261.

   https://doi.org/10.1038/nrg3414

   • Google Scholar
6. 1. Ezkurdia I
   2. Graña O
   3. Izarzugaza JM
   4. Tress ML

   (2009) Assessment of domain boundary predictions and the prediction of intramolecular contacts in CASP8

   Proteins 77:196–209.

   https://doi.org/10.1002/prot.22554

   • Google Scholar
7. 1. Faure G
   2. Bornot A
   3. de Brevern AG

   (2008) Protein contacts, inter-residue interactions and side-chain modelling

   Biochimie 90:626–639.

   https://doi.org/10.1016/j.biochi.2007.11.007

   • Google Scholar
8. 1. Finn RD
   2. Bateman A
   3. Clements J
   4. Coggill P
   5. Eberhardt RY
   6. Eddy SR
   7. Heger A
   8. Hetherington K
   9. Holm L
   10. Mistry J
   11. Sonnhammer EL
   12. Tate J
   13. Punta M

   (2014) Pfam: the protein families database

   Nucleic Acids Research 42:D222–D230.

   https://doi.org/10.1093/nar/gkt1223

   • Google Scholar
9. 1. Fodor AA
   2. Aldrich RW

   (2004) Influence of conservation on calculations of amino acid covariance in multiple sequence alignments

   Proteins 56:211–221.

   https://doi.org/10.1002/prot.20098

   • Google Scholar
10. 1. Gloor GB
    2. Martin LC
    3. Wahl LM
    4. Dunn SD

    (2005) Mutual information in protein multiple sequence alignments reveals two classes of co-evolving positions

    Biochemistry 44:7156–7165.

    https://doi.org/10.1021/bi050293e

    • Google Scholar
11. 1. Göbel U
    2. Sander C
    3. Schneider R
    4. Valencia A

    (1994) Correlated mutations and residue contacts in proteins

    Proteins 18:309–317.

    https://doi.org/10.1002/prot.340180402

    • Google Scholar
12. 1. Grant BJ
    2. Rodrigues AP
    3. ElSawy KM
    4. McCammon JA
    5. Caves LS

    (2006) Bio3d: an R package for the comparative analysis of protein structures

    Bioinformatics 22:2695–2696.

    https://doi.org/10.1093/bioinformatics/btl461

    • Google Scholar
13. 1. Horovitz A
    2. Bochkareva ES
    3. Yifrach O
    4. Girshovich AS

    (1994) Prediction of an inter-residue interaction in the chaperonin GroEL from multiple sequence alignment is confirmed by double-mutant cycle analysis

    Journal of Molecular Biology 238:133–138.

    https://doi.org/10.1006/jmbi.1994.1275

    • Google Scholar
14. 1. Jones DT
    2. Buchan DW
    3. Cozzetto D
    4. Pontil M

    (2012) PSICOV: precise structural contact prediction using sparse inverse covariance estimation on large multiple sequence alignments

    Bioinformatics 28:184–190.

    https://doi.org/10.1093/bioinformatics/btr638

    • Google Scholar
15. 1. Kamisetty H
    2. Ovchinnikov S
    3. Baker D

    (2013) Assessing the utility of coevolution-based residue-residue contact predictions in a sequence- and structure-rich era

    Proceedings of the National Academy of Sciences of USA 110:15674–15679.

    https://doi.org/10.1073/pnas.1314045110

    • Google Scholar
16. 1. Kass I
    2. Horovitz A

    (2002) Mapping pathways of allosteric communication in GroEL by analysis of correlated mutations

    Proteins 48:611–617.

    https://doi.org/10.1002/prot.10180

    • Google Scholar
17. 1. Kersey PJ
    2. Allen JE
    3. Christensen M
    4. Davis P
    5. Falin LJ
    6. Grabmueller C
    7. Hughes DS
    8. Humphrey J
    9. Kerhornou A
    10. Khobova J
    11. Langridge N
    12. McDowall MD
    13. Maheswari U
    14. Maslen G
    15. Nuhn M
    16. Ong CK
    17. Paulini M
    18. Pedro H
    19. Toneva I
    20. Tuli MA
    21. Walts B
    22. Williams G
    23. Wilson D
    24. Youens-Clark K
    25. Monaco MK
    26. Stein J
    27. Wei X
    28. Ware D
    29. Bolser DM
    30. Howe KL
    31. Kulesha E
    32. Lawson D
    33. Staines DM

    (2015) Ensemblegeneomes

    Ensemblegeneomes, http://ensemblgenomes.org/.

    http://ensemblgenomes.org/

    • Google Scholar
18. 1. Lee J
    2. Natarajan M
    3. Nashine VC
    4. Socolich M
    5. Vo T
    6. Russ WP
    7. Benkovic SJ
    8. Ranganathan R

    (2008) Surface sites for engineering allosteric control in proteins

    Science 322:438–442.

    https://doi.org/10.1126/science.1159052

    • Google Scholar
19. 1. Leinonen R
    2. Akhtar R
    3. Birney E
    4. Bower L
    5. Cerdeno-Tárraga A
    6. Cheng Y
    7. Cleland I
    8. Faruque N
    9. Goodgame N
    10. Gibson R
    11. Hoad G
    12. Jang M
    13. Pakseresht N
    14. Plaister S
    15. Radhakrishnan R
    16. Reddy K
    17. Sobhany S
    18. Ten Hoopen P
    19. Vaughan R
    20. Zalunin V
    21. Cochrane G

    (2015) European Nucleotide Archive

    European Nucleotide Archive, EBI ENA, http://www.ebi.ac.uk/ena.

    http://www.ebi.ac.uk/ena

    • Google Scholar
20. 1. Livesay DR
    2. Kreth KE
    3. Fodor AA

    (2012) A critical evaluation of correlated mutation algorithms and coevolution within allosteric mechanisms

    Methods in Molecular Biology 796:385–398.

    https://doi.org/10.1007/978-1-61779-334-9_21

    • Google Scholar
21. 1. Mao W
    2. Kaya C
    3. Dutta A
    4. Horovitz A
    5. Bahar I

    (2015) Comparative study of the effectiveness and limitations of current methods for detecting sequence coevolution

    Bioinformatics 31:1929–1937.

    https://doi.org/10.1093/bioinformatics/btv103

    • Google Scholar
22. 1. Marks DS
    2. Hopf TA
    3. Sander C

    (2012) Protein structure prediction from sequence variation

    Nature Biotechnology 30:1072–1080.

    https://doi.org/10.1038/nbt.2419

    • Google Scholar
23. 1. McWilliam H
    2. Valentin F
    3. Goujon M
    4. Li W
    5. Narayanasamy M
    6. Martin J
    7. Miyar T
    8. Lopez R

    (2009) Web services at the European Bioinformatics Institute-2009

    Nucleic Acids Research 37:W6–W10.

    https://doi.org/10.1093/nar/gkp302

    • Google Scholar
24. 1. Morcos F
    2. Pagnani A
    3. Lunt B
    4. Bertolino A
    5. Marks DS
    6. Sander C
    7. Zecchina R
    8. Onuchic JN
    9. Hwa T
    10. Weigt M

    (2011) Direct-coupling analysis of residue coevolution captures native contacts across many protein families

    Proceedings of the National Academy of Sciences of USA 108:E1293–E1301.

    https://doi.org/10.1073/pnas.1111471108

    • Google Scholar
25. 1. Noivirt O
    2. Eisenstein M
    3. Horovitz A

    (2005) Detection and reduction of evolutionary noise in correlated mutation analysis

    Protein Engineering, Design & Selection 18:247–253.

    https://doi.org/10.1093/protein/gzi029

    • Google Scholar
26. 1. Pollock DD
    2. Taylor WR
    3. Goldman N

    (1999) Co-evolving protein residues: maximum likelihood identification and relationship to structure

    Journal of Molecular Biology 287:187–198.

    https://doi.org/10.1006/jmbi.1998.2601

    • Google Scholar
27. 1. Pruitt KD
    2. Brown GR
    3. Hiatt SM
    4. Thibaud-Nissen F
    5. Astashyn A
    6. Ermolaeva O
    7. Farrell CM
    8. Hart J
    9. Landrum MJ
    10. McGarvey KM
    11. Murphy MR
    12. O'Leary NA
    13. Pujar S
    14. Rajput B
    15. Rangwala SH
    16. Riddick LD
    17. Shkeda A
    18. Sun H
    19. Tamez P
    20. Tully RE
    21. Wallin C
    22. Webb D
    23. Weber J
    24. Wu W
    25. Dicuccio M
    26. Kitts P
    27. Maglott DR
    28. Murphy TD
    29. Ostell JM

    (2015) RefSeq

    RefSeq, NCBI Reference sequence database, http://www.ncbi.nlm.nih.gov/refseq.

    http://www.ncbi.nlm.nih.gov/refseq

    • Google Scholar
28. 1. Rice P
    2. Longden I
    3. Bleasby A

    (2000) EMBOSS: the European molecular biology open software suite

    Trends in Genetics 16:276–277.

    https://doi.org/10.1016/S0168-9525(00)02024-2

    • Google Scholar
29. 1. Shilton BH
    2. Thomas DY
    3. Cygler M

    (1997) Crystal structure of Kex1Δp, a prohormone-processing carboxypeptidase from Saccharomyces cerevisiae

    Biochemistry 36:9002–9012.

    https://doi.org/10.1021/bi970433n

    • Google Scholar
30. 1. Skwark MJ
    2. Raimondi D
    3. Michel M
    4. Elofsson A

    (2014) Improved contact predictions using the recognition of protein like contact patterns

    PLOS Computational Biology 10:e1003889.

    https://doi.org/10.1371/journal.pcbi.1003889

    • Google Scholar
31. 1. Wang G
    2. Dunbrack RL Jr

    (2003) PISCES: a protein sequence culling server

    Bioinformatics 19:1589–1591.

    https://doi.org/10.1093/bioinformatics/btg224

    • Google Scholar
32. 1. Weigt M
    2. White RA
    3. Szurmant H
    4. Hoch JA
    5. Hwa T

    (2009) Identification of direct residue contacts in protein-protein interaction by message passing

    Proceedings of the National Academy of Sciences of USA 106:67–72.

    https://doi.org/10.1073/pnas.0805923106

    • Google Scholar
33. 1. Wollenberg KR
    2. Atchley WR

    (2000) Separation of phylogenetic and functional associations in biological sequences by using the parametric bootstrap

    Proceedings of the National Academy of Sciences of USA 97:3288–3291.

    https://doi.org/10.1073/pnas.070154797

    • Google Scholar
34. 1. Yates A
    2. Beal K
    3. Keenan S
    4. McLaren W
    5. Pignatelli M
    6. Ritchie GR
    7. Ruffier M
    8. Taylor K
    9. Vullo A
    10. Flicek P

    (2015) The Ensembl REST API: ensembl data for any language

    Bioinformatics 31:143–145.

    https://doi.org/10.1093/bioinformatics/btu613

    • Google Scholar

Author details

1. Etai Jacob

   1. The Mina and Everard Goodman Faculty of Life Sciences, Bar-Ilan University, Ramat-Gan, Israel
   2. Department of Structural Biology, Weizmann Institute of Science, Rehovot, Israel

   Contribution

   EJ, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

2. Ron Unger

   The Mina and Everard Goodman Faculty of Life Sciences, Bar-Ilan University, Ramat-Gan, Israel

   Contribution

   RU, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   ron@biomodel.os.biu.ac.il

   Competing interests

   The authors declare that no competing interests exist.

3. Amnon Horovitz

   Department of Structural Biology, Weizmann Institute of Science, Rehovot, Israel

   Contribution

   AH, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   Amnon.Horovitz@weizmann.ac.il

   Competing interests

   The authors declare that no competing interests exist.

• 3,135

  views

• 604

  downloads

• 8

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.