PYCOEVOL

A Python Workflow to Study Protein-protein Coevolution

Fábio Madeira and Ludwig Krippahl

CENTRIA-DI, Universidade Nova de Lisboa, Caparica, Portugal

Keywords: Protein coevolution, Multiple sequence alignments, Mutual information.

Abstract: Protein coevolution has emerged as an important research topic. Several methods and scoring systems were

developed to quantify coevolution, though the quality of the results usually depends on the completeness of

the biological data. To simplify the computation of coevolution indicators from the data, we have

implemented a fully integrated and automated workflow which enables efficient analysis of protein

coevolution, using the Python scripting language. Pycoevol automates access to remote or local databases

and third-party applications, including also data processing functions. For a given protein complex under

study, Pycoevol retrieves and processes all the information needed to undergo the analysis, namely

homologous sequence search, multiple sequence alignment computation and coevolution analysis, using a

Mutual Information indicator. In addition, friendly output results are created, namely histograms and

heatmaps of inter-protein mutual information scores, as well as lists of significant coevolving residue pairs.

An illustrative example is presented. Pycoevol is platform independent, and is available under the general

public license from http://code.google.com/p/pycoevol.

1 INTRODUCTION

Protein coevolution has emerged as an important

research topic, being applied successfully to predict

the structure of RNA (Freyhult et al., 2005) and

proteins (Yeang and Haussler, 2007); to predict

intermolecular interactions (Pazos et al., 1997); to

identify functionally important regions of molecules

(Saraf et al., 2003); and to identify energetic

pathways through molecules (Süel et al., 2003).

Overall, protein coevolution corresponds to the

accumulation of structural/functional changes

through evolutionary lineages, which are

compensated by changes in other regions of the

same protein or in another protein (Pazos and

Valencia, 2008). Since many proteins have evolved

performing mutual interactions and consequently

forming specific molecular complexes (Pazos and

Valencia, 2008), inter-protein coevolution

corresponds to mutual evolutionary constraints

imposed by each protein on the other partner and

accordingly, protein sequences must reflect this

evolutionary process (Pazos et al., 1997).

Coevolving residues are detected in a three-step

process: 1) search for homologue sequences; 2)

computation of a multiple sequence alignment

(MSA) for each protein; 3) calculation of a

coevolution score for each pair of sites in the MSAs.

The first step involves the selection of orthologues

and the matching of the correct protein pairs along

organism lineages. One approach can be a PSI-

BLAST search (Altschul et al., 1997), which is a

reliable source for distant relatives and favours

orthologues over paralogues, reducing one source of

errors. The second step involves the computation of

MSAs for each protein. The MSA is a rich source of

sequence-function relationships and attempts to

represent the evolutionary relations between the

homologous sequences by aligning them under the

assumption that mutations are independent.

Although the MSA relies in computationally

complex algorithms (Elias, 2006), considering the

number and the variability of a set of homologous

sequences, supplemented with the assumed

independence of mutations, this often leads to poor

alignments, and becomes an important problem for

identifying coevolution. Finally, a coevolution score

is calculated for each pair of sites in the MSAs.

Coevolution analysis depends on the

completeness of the biological data and, given the

complexity of MSA computations, sometimes made

worse by insufficient evolutionary divergence and

143

Madeira F. and Krippahl L..

PYCOEVOL - A Python Workﬂow to Study Protein-protein Coevolution.

DOI: 10.5220/0003737901430149

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (BIOINFORMATICS-2012), pages 143-149

ISBN: 978-989-8425-90-4

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

the presence of too many indels (gaps), a large

number of methods and scoring functions have been

proposed in the literature to estimate protein

coevolution, including phylogenetic independent and

phylogenetic dependent methods (see (Halperin et

al., 2006) and (Caporaso et al., 2008) for extended

reviews). It is likely that different applications

would require different methods and it can be

difficult to choose from them, as they exhibit subtle

yet significant differences. Unfortunately, there isn’t

any survey or benchmark assessing the accuracy of

each approach.

Mutual Information (MI) (Martin et al., 2005) is

a standard measure to identify sites of correlated and

compensatory mutations in a set of homologous

sequences and can be used to identify correlated

positions between two interacting proteins. MI is a

measure based on Information Theory, which

quantifies the mutual dependence between two

random variables, and is a “tree-ignorant” method

(Caporaso et al., 2008), since it does not account for

likelihoods or shared ancestral correlations. In a

MSA, the MI between two columns (positions)

reflects the extent to which the existence of one

specific amino acid residue at one position allows us

to predict the identity of the residue at the other

position. MI scores are high if substitutions at the

two positions show positive correlation.

Since protein-protein coevolution analysis

involves the execution of the specified steps, and for

each step there is a multitude of options, we have

implemented a fully integrated and automated

workflow which combines these steps and enables

efficient analysis of coevolution among proteins,

using the Python scripting language. Pycoevol aims

at automating the computation of coevolution

indicators, speeding up data-mining and improving

the accuracy of the results. Pycoevol automates

access to remote or local databases and third-party

applications and includes data processing functions,

thus enabling a flexible use of the workflow.

As a proof-of-concept we present the complete

analysis of coevolution in a protein complex formed

by transforming growth factor beta 3 (TGF-β3) and

extracellular domain of TGF-β receptor type II

(TGF-β receptor) (Hart et al., 2002). TGF-β3 is a

protein that controls cellular proliferation and

differentiation by signalling through kinase

receptors, namely serine/threonine receptors as

TGF-β receptor. The execution of the Pycoevol

workflow pinpointed several coevolving pairs of

residues. Most residues identified were at the surface

of the protein complex, and 13% of them were

located at the complex interface. This example

further illustrates how computing inter-protein

coevolution can improve the accuracy of constrained

docking algorithms (e.g. BiGGER (Palma et al.,

2000)) assisting the demanding task of protein

docking.

2 METHODS

As described in the previous section, protein

coevolution analysis involves the execution of a

series of steps, and for each step there is a multitude

of options. To congregate these steps in a system

which enables an efficient analysis of protein

coevolution, we developed a Python workflow

consisting of a set of scripts, which includes

connectors to local or remote databases, enabling

also the execution of third party applications through

the command line, for both PSI-Blast search and

computation of MSAs. Furthermore, we have used

the Biopython module (Cock et al., 2009), for

general manipulation of biological data.

The analysis starts with the input of Protein Data

Bank accession numbers (PDB ID), for each protein

partner. Alternatively, accession numbers for NCBI

reference sequence identifiers (GI) or UniProt

primary (citable) accession number can be also used,

as in the case of proteins without available 3D

structures. For each protein a collection of

orthologous sequences is searched. This is done by a

PSI-BLAST search (Altschul et al., 1997) against

NCBI Reference Proteins database. The reason for

using PSI-BLAST instead of a simple BLAST

search is that we need to start with a large number of

ortologous sequences and PSI-BLAST allows us to

find more distant relatives. PSI-blast uses the default

configuration, but alternative configurations can be

also specified. Protein sequences of both partners are

then matched by comparing the source organisms for

each sequence. The assumption is that, if the

proteins are homologous and interact in one

organism, they should also interact in the other

organisms. This is a crucial step because proteins

will only coevolve within the same lineage, and

without matching the correct organisms the data

obtained would be meaningless. A refined set of

sequences for each interacting partner is obtained

and three different MSAs are computed, using

ClustalW (Chenna et al., 2003) and Muscle (Edgar,

2004) in the default configuration, and Mafft (Katoh

et al., 2002) with linsi (L-INS-i), the most accurate

configuration. In order to compare the performance

of the MSAs computed, we develop two scores: SP

score, which scores each MSA (GOP 4.0, GEP 1.0

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

144

and the BLOSUM62 scoring matrix (Henikoff and

Henikoff, 1992)); and a column score (CS), which

compares the MSAs column by column. These

scoring systems are implemented in Python and are

based on the ones developed for BaliBASE

(Thompson et al., 1999). Instead of comparing the

SP scores against a reference MSA, as in the case of

BaliBASE approach, our implementation compares

the scores obtained for each MSA computed with

different MSA programs. For the MSAs

corresponding to each protein partner, the ones with

the best scores are selected to further analysis.

Supplementing the SP, CS gives an overview on the

similarity between MSAs, and the higher the CS; the

higher the percentage of equal columns present on

both MSAs. The selected MSAs are also inspected

for misalignments and for specific destabilizing

sequences, and are cropped by excluding portions

not covered by the structures if a 3D structure is

available, otherwise the complete sequences are

used. The next step consists in estimating which

pairs of residues show traces of coevolution. The

MSA columns represent a snapshot of the

evolutionary relations of all different protein

sequences for that position. The estimation is

calculated for all pairs of columns between two

MSAs, by Mutual Information (MI), which is

defined as follows (Martin et al., 2005):

MI (X,Y) = H(X) + H(Y) - H(X,Y) (1)

The estimation is done summing the entropies H(X)

and H(Y), corresponding to amino acid frequencies

for each residue belonging to the columns in

analysis, minus the joint entropy H(X,Y) for that

pair of columns. The higher the estimation is, the

higher the indication that compensatory mutations

arose that mitigated the deleterious effects of

mutations interfering with the stability of the

interaction between that specific pair of residues.

Estimation of MI (1) for all pairs of columns in both

MSAs gives a matrix of scores and, to improve the

detection of significant positions, the analysis is

constrained to include only the residues belonging to

the surface of the proteins. The assumption is that

residues on the core of the protein are not

susceptible to perform or sense the influence of

residues at the interface of the other protein. We

implemented a measure of accessible surface area

(ASA) based on the surface calculations

implemented in Hollow (Ho and Gruswitz, 2008).

Alternatively, we implemented a measure of solvent

excluded surfaces of molecules, using the MSMS

program (Sanner et al., 1996). In any case, for each

position in the MSAs there are a multitude of

contacts and the results of this estimations show that

for each residue in one protein the possible mutual

propensity contacts between positions are usually

variable.

For each complex, the map of potential

interactions is calculated using a MI calculator, and

friendly output results are created, namely

histograms and heatmaps of inter-protein mutual

information scores, as well as lists of significant

coevolving residue pairs. From the collection of MI

scores (surface positions), only the best 20% scores

are selected as significant coevolving residues, and

are prone to further analysis.

3 RESULTS AND DISCUSSION

Pycoevol detects inter-protein coevolution, being

therefore suitable to the study of protein complexes.

Furthermore, this workflow is optimized to predict

intermolecular interactions and to constraint the

search space in protein docking. It can be also used

to complement studies concerning the identification

of functionally and structurally important regions of

molecules as well as to predict the structure of

proteins. It offers an easy implementation of the

protein-protein coevolution analysis workflow

(Figure 1), which congregates several tools and

tasks, enabling the researcher to focus on biological

problems, rather than repetitive execution or

implementation of self-made codes. Only if special

tools or specific tasks are needed, the user may need

to edit or complement the Pycoevol source code. For

the general user, Pycoevol runs like any other

application or any other python script, called from

the command line. The simplest usage is to type

python pycoevol.py at the prompt of the operating

system’s command line and select from the available

options. The execution of the workflow takes from

minutes to hours, depending on the extension of the

data being processed. To analyse the traces of

coevolution between the proteins in the complex, the

user only have to specify the accession numbers of

the protein sequences. After PSI-BLAST search,

MSA computation and coevolution analysis, easy

readable output results are generated.

As we noticed before, several methods and

scoring systems were developed to detect traces of

coevolution, though the quality of the results also

depends on the quality of the source data.

Accordingly, the detection of inter-protein

coevolution relies on two main aspects. One is the

computation of coevolution indicators and, to

improve our application, we plan to include other

PYCOEVOL - A Python Workflow to Study Protein-protein Coevolution

145

coevolution indicators as well as normalizations,

increasing therefore the sensibility and allowing the

cross-validation of the most significant results.

Figure 1: Main steps implemented in the Pycoevol

workflow.

The other important factor is the quality of the

data and MSA being processed, so we plan to

include other MSA programs and a refining tool that

accounts for the dependence of mutations, thus

improving the quality of the MSAs. Beyond the

advantages to the user, these improvements will

eventually let us perform a survey assessing the

accuracy of some coevolution indicators, as well as

evaluate which MSA program is better suited to

protein-protein coevolution analysis. T-Coffee

Expresso (Notredame et al., 2000) is a good

candidate to be included in our workflow, as it

produces structural alignments, rather than sequence

alignments, and is a powerful and accurate

alignment tool based on the construction of libraries

(local alignments), which then incorporates

additional structural information to obtain the final

MSA. In theory, the advent of structural information

to constraint and to compute MSAs should be a

more realistic source of information towards

detection of inter-protein coevolution, since residues

are aligned based on structural topologies rather than

just to maximize a scoring function, as in the case of

the classic Sum-of-Pairs (SP) scoring system usually

implemented in MSA algorithms (Do and Katoh,

2008). Although the details of which scoring

function is better suited to co-evolutionary analysis

or how to infer inter-molecular contacts from MSA

data requires serious benchmarking, our workflow

uses three standard MSA programs with different

performances to compute the alignments. The MSA

programs included are the widely used ClustalW and

Muscle, and Mafft, ranked the best program in

overall performance in a recent benchmark based on

the BaliBASE database (Thompson et al., 2011).

The comparison between the SP scores of each

MSA, gives a preliminary measure on what MSA

program may favour the detection of coevolution for

the specific test case under study, accounting on the

principle that an MSA with high SP score is better

suited to MI analysis. This can be misleading, since

the MSA algorithm can misplace coevolving sites in

the alignment, while attempting to maximize a

scoring function (as the SP score).

Pycoevol was tested on the protein complex

TGF-β3/TGF-β receptor. Figure 2A shows a

histogram of MI score frequencies computed for all

the pairs of positions on the MSAs computed for

each protein. In this case, coevolution results were

processed using MSAs computed by Mafft, as they

scored best. Along with histograms, heatmaps were

also generated (Figure 2B and 2C). These give a

general view on the relative positions of highly

scoring residues belonging to each protein. By

selecting only the highest MI scores, the heatmap

gives a clearer view on the positions of those

residues. The structural analysis of the protein

complex shows the location of interface residues for

both protein partners (Figure 2D). Most residues

presenting higher MI scores were located at the

surface (Figure 2E bottom), and 13% of them (3 out

of 23) were located at the interface (Figure 2E top).

The interacting map obtained for the TGF-β3/TGF-β

receptor complex (Figure 2F), i.e. the network of

pairings relatives to the highest scoring positions,

shows that TGF-β3 has 2 residues on the interface (2

out of 8) and TGF-β receptor has 1 residue on the

interface (1 out of 15). This finding is very

interesting; given the complexity of protein-protein

docking, finding even only one positive interface

contact can help constraint the search space and

improve the accuracy of constrained docking

algorithms.

To test the application of our coevolution measure-

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

146

Figure 2: Protein coevolution analysis of complex TGF-β3/TGF-β receptor with mutual information. Every pair of positions

on the MSAs was examined for coevolution using an analysis of MI. (A) Histogram of inter-protein mutual information

scores. (B) The matrix of scores for each pair of positions in the alignment is plotted as a heatmap, according to the colour

legend shown. (C) The matrix of scores for positions with highest MI scores (greater than 80% of the best MI score) is

plotted as a heatmap. (D) Illustration of complex TGF-β3/TGF-β receptor (PDB ID 1KTZ) showing interface residues of

TGF-β3 as blue spheres (top) and interface residues of TGF-β receptor as red spheres (bottom). (E) Illustration of

coevolving residues of TGF-β3 (blue spheres) and TGF-β receptor (red spheres), located at the interface (top). Previous

representation with remaining coevolving residues of TGF-β3 (light blue spheres) and TGF-β receptor (orange spheres),

located at the surface (bottom). (F) Network of the pairings identified by MI, using the same colour scheme as in the

previous panel (E).

ments on protein docking, we employed the

coevolving residues identified on the complex TGF-

β3/TGF-β receptor, as constraints in BiGGER. This

is a difficult complex to model because of the low

contact surface. Using unbound structures of both

TGF-β3 and TGF-β receptor (PDB structures 1TGK

and 1MZ9, respectively), to simulate a real-life

application, BiGGER could not retain, within the

5000 highest scoring surface contacts, any model

that placed the TGF-β receptor at the correct

position (Figure 3A). This was no longer the case

when constraining the docking simulation using

coevolution data. From TGF-β3 we considered 8

residues with higher MI score, in our coevolution

criteria, two of which were truly at the interface.

From TGF-β receptor there were 15 residues that

fulfilled our selection criteria, with only one at the

interface. One approach was to use both sets of

residues, but imposing the constraint that there must

be at least one contact between the 8 residues of one

partner and the 15 of the other. Contact was defined

as having the alpha Carbons at most 7Å apart. Panel

B of Figure 3 shows one model obtained as a result,

with TGF-β receptor placed at the correct point

relative to TGF-β3, though not quite at the right

orientation. This experiment shows that even

without being able to identify exactly which are the

correct contacts, the constraint requiring any one

contact, at least, reduces the search space enough so

that correct models are not lost during the filtering

stage of docking. In Figure 3C we show one model

obtained by forcing the specific contact between an

interface residue of TGF-β3 (A93) and one interface

residue of TGF-β receptor (R52). Since both

residues are at the interface, all models are

approximately at the right position. The interesting

aspect of this experiment is that, with this constraint,

the search space is so reduced that a docking takes

less than five minutes, as opposed to nearly two

hours for an unconstrained docking run. This means

that it is feasible to test, individually, all potential

residue contacts given by the coevolution measure-

PYCOEVOL - A Python Workflow to Study Protein-protein Coevolution

147

Figure 3: Summary of the docking results. (A) Correct structure of the complex TGF-β3/TGF-β receptor (PDB ID 1TGK

and 1M9Z, respectively), surrounded by spheres representing the geometric centre of TGF-β receptor (red), where it was

predicted in the 5000 models generated in the docking run. No model was found that places TGF-β receptor at the right

position. (B) One of the models obtained by requiring an unspecified contact between the two groups of 8 and 15 candidate

interface residues. In both cases, the resulting model was not very accurate, given the incorrect orientation of TGF-β

receptor (red), but even so this was a significant improvement over the unconstrained docking in the modelling of this

complex. (C) One of the models obtained by restricting the distance between residue A93 of TGF-β3(blue) and residue R52

of TGF-β receptor (red), which run took less than five minutes.

ments. Furthermore, since the docking runs are

independent, the whole process is trivial to run in

parallel, thus, in practice, taking less time than a

single unconstrained docking run.

4 CONCLUSIONS

Pycoevol source code (version 1.0) will be made

freely available for downloaded from

http://code.google.com/p/pycoevol.

Additional information on the third-party

dependencies, as well as how to install and run the

program can be check at the same location. The

workflow is fully written in Python 2.7, platform

independent, and is available under the general

public license. User requests and contributions may

be implemented and included in future versions of

the software.

Pycoevol is a fully integrated and automated

system which enables efficient analysis of

coevolution among proteins. In order to improve the

workflow and its capabilities, we plan to include

other coevolution indicators as well as

normalizations. Alongside, the inclusion of new

MSA programs and a refining tool is also planned.

The integration with molecular viewers, such as

RasMol (Bernstein, 2000) and UCSF-Chimera

(Pettersen et al., 2004), which allows the automatic

selection and display of the coevolving residues, will

enable further analysis and usage possibilities.

Finally, the natural extension of the Pycoevol

workflow aims at include also protein docking and

consequently, Pycoevol will be included and

integrated in the Open Chemera Library (Krippahl,

2011), an open source library, which includes among

several features, BiGGER, the constrained docking

algorithm. This system will become a useful tool to

study protein coevolution and interaction, and will

hopefully be the basis for several studies concerning

protein interactions.

ACKNOWLEDGEMENTS

This work was funded by national funds from FCT-

Fundação para a Ciência e Tecnologia, under project

PTDC/EIA-CCO/115999/2009.

REFERENCES

Altschul, S. F., Madden, T. L., Schäffer, A. A., Zhang, J.,

Zhang, Z., Miller, W., Lipman, D. J. (1997). Gapped

BLAST and PSI-BLAST: a new generation of protein

database search programs. Nucleic Acids Research,

25(17), 3389-402.

Bernstein, H. J. (2000). Recent changes to RasMol,

recombining the variants. Trends in Biochemical

Sciences, 25(9), 453-5.

Caporaso, J. G., Smit, S., Easton, B. C., Hunter, L.,

Huttley, G. A., Knight, R. (2008). Detecting

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

148

coevolution without phylogenetic trees? Tree-ignorant

metrics of coevolution perform as well as tree-aware

metrics. BMC Evolutionary Biology, 8, 327-52.

Chenna, R., Sugawara, H., Koike, T., Lopez, R., Gibson,

T. J., Higgins, D. G., Thompson, J. D. (2003).

Multiple sequence alignment with the Clustal series of

programs. Nucleic Acids Research, 31(13), 3497-500.

Cock, P. J., Antao, T., Chang, J. T., Chapman, B., Cox, C.

J., Dalke, A., Friedberg, I., Hamelryck, T., Kauff, F.,

Wilczynski, B. (2009). Biopython: freely available

Python tools for computational molecular biology and

bioinformatics. Bioinformatics, 25(11), 1422-3.

Do, C. B., Katoh, K. (2008). Protein Multiple Sequence

Alignment. Functional Proteomics: Methods and

Protocols, 484, 379-413.

Edgar, R. C. (2004). MUSCLE: multiple sequence

alignment with high accuracy and high throughput.

Nucleic Acids Research, 32(5), 1792-7.

Elias, I. (2006). Settling the intractability of multiple

alignment. Journal of Computational Biology, 13(7),

1323-39.

Freyhult, E., Moulton, V., Gardner, P. (2005). Predicting

RNA structure using mutual information. Applied

Bioinformatics, 4(1), 53-59.

Halperin, I., Wolfson, H., Nussinov, R. (2006). Correlated

Mutations: Advances and Limitations. A Study on

Fusion Proteins and on the Cohesin-Dockerin

Families. Proteins: Structure, Function, and

Bioinformatics, 63(4), 832–845.

Hart, P. J., Deep, S., Taylor, A. B., Shu, Z., Hinck, C. S.,

Hinck, A. P. (2002). Crystal structure of the human

TbetaR2 ectodomain--TGF-beta3 complex. Nature

Structural Biology, 9(3), 203-8.

Henikoff, S., Henikoff, J.G. (1992). Amino acid

substitution matrices from protein blocks. Proceedings

of the National Academy of Sciences of the United

States of America, 89(22), 10915-9.

Ho, B. K., Gruswitz, F. (2008). HOLLOW: generating

accurate representations of channel and interior

surfaces in molecular structures. BMC Structural

Biology, 8, 49-55.

Katoh, K., Misawa, K., Kuma, K., Miyata, T. (2002).

MAFFT: a novel method for rapid multiple sequence

alignment based on fast Fourier transform. Nucleic

Acids Research, 30(14), 3059-66.

Krippahl, L. (2011). Open Chemera Library. Available at

https://github.com/lkrippahl/Open-Chemera

Martin, L. C., Gloor, G. B., Dunn, S. D., Wahl, L. M.

(2005). Using information theory to search for co-

evolving residues in proteins. Bioinformatics, 21(22),

4116-24.

Notredame, C., Higgins, D. G., Heringa, J. (2000). T-

Coffee: A novel method for fast and accurate multiple

sequence alignment. Journal of Molecular Biology,

302(1), 205-17.

Palma, P. N., Krippahl, L., Wampler, J. E., Moura, J. J.

(2000). BiGGER: a new (soft) docking algorithm for

predicting protein interactions. Proteins, 39(4), 372-

84.

Pazos, F., Helmer-Citterich, M., Ausiello, G., Valencia, A.

(1997). Correlated mutations contain information

about protein-protein interaction. Journal of Molecular

Biology, 271(4), 511-23.

Pazos, F.,Valencia, A. (2008). Protein co-evolution, co-

adaptation and interactions.

The EMBO Journal,

27(20), 2648-55.

Pettersen, E. F., Goddard, T. D., Huang, C. C., Couch, G.

S., Greenblatt, D. M., Meng, E. C., Ferrin, T. E.

(2004). UCSF Chimera--a visualization system for

exploratory research and analysis. Journal of

Computional Chemistry, 25(13), 1605-13.

Sanner, M. F., Olson, J. Spehner, J. C. (1996). Reduced

surface: an efficient way to compute molecular

surfaces. Biopolymers, 38(3), 305-20.

Saraf, M., Moore, G., Maranas, C. (2003). Using multiple

sequence correlation analysis to characterize

functionally important protein regions. Protein

Engineering, 16(6), 397-406.

Süel, G. M., Lockless, S. W., Wall, M. A., Ranganathan,

R. (2003). Evolutionarily conserved networks of

residues mediate allosteric communication in proteins.

Nature Structural Biology, 10(1), 59-69.

Thompson, J. D., Plewniak, F., Poch, O. (1999).

BAliBASE: a benchmark alignment database for the

evaluation of multiple alignment programs.

Bioinformatics, 15(1), pp.87-8.

Thompson, J. D., Linard, B., Lecompte, O., Poch, O.

(2011). A Comprehensive Benchmark Study of

Multiple Sequence Alignment Methods: Current

Challenges and Future Perspectives. PLoS ONE, 6(3),

18093-107.

Yeang, C-H., Haussler, D. (2007). Detecting coevolution

in and among protein domains. PLoS Computational

Biology, 3(11), 2122-34.

PYCOEVOL - A Python Workflow to Study Protein-protein Coevolution

149