Protein Homology Modeling
Manuel C Peitsch, Novartis Pharma AG and Swiss Institute of Bioinformatics,
Basel, Switzerland
Torsten Schwede, Biozentrum, Universita¨t Basel and Swiss Institute of Bioinformatics,
Basel, Switzerland
Alexander Diemand, Glaxo Wellcome Experimental Research and Swiss Institute of
Bioinformatics, Geneva, Switzerland
Nicolas Guex, Glaxo Wellcome Experimental Research and Swiss Institute of Bioinformatics,
Geneva, Switzerland
Protein homology modeling is the prediction of the three-dimensional structure of proteins
by comparative methods that use the known three-dimensional structure of related
proteins. This method will play a major role in the functional analysis of the genes (and their
protein transcripts) discovered in fully sequenced genomes.
Introduction
Understanding the function and physiological role of
proteins is a basic requirement for the discovery of
novel medicines (small molecules) and ‘biologicals’
(protein-based products) with medical, industrial or
commodity applications. Although the sequence of the
human genome has been deciphered, we are very far
from understanding the function and the physiological
role of the gene products it encodes. Indeed, being able
to read the letters and the words is not connected with
understanding their meaning. Therefore, the attention
of many biologists is now shifting to functional
genomics which aims to discover the function and
physiological role of the gene products encoded in the
genome. Functional genomics is a very complex field
and requires a combination of technologies. Consequently,
new experimental approaches, and their
automation for large-scale applications, will need
development. Concurrently, and in order to maximize
the value of large data sets, one will witness the
development of new data mining methods and
mathematical models for the simulation of biological
processes. One area where this development has
started is structural genomics, or the elucidation of
the three-dimensional (3D) structure of proteins
discovered in the genome sequences. All the major
steps in experimental protein structure determination
will be optimized and automated as much as possible.
Protein homology modeling will allow for the rapid
generation of new protein models based on this large
body of newly elucidated structures, effectively increasing
the impact of these research programs.
A protein’s function is tightly linked to its 3D
structure. As residues located far apart in the primary
sequence can be very close in space, and only a few
residues are generally responsible for a protein’s
function, insights into the 3D structure of a protein
can represent a key component of the process of
functional analysis. Consequently, an atomic-level 3D
representation to assign roles to specific residues is a
major asset, both for planning experiments and for
explaining observations.
The ‘folding’ process of a protein is very complex,
and no objective and reliable way to determine it from
the sequence alone has as yet been developed.
Scientists are thus dependent on experimental elucidation
of protein structure. The usual approaches,
both X-ray diffraction and nuclear magnetic resonance
(NMR), are, however, hampered by many
technical hurdles and limitations. Consequently, several
concerted ‘structural genomics’ efforts are being
launched in both the private and public sector to
address these difficulties and increase the throughput
of experimental elucidation of structure. However,
these efforts will not be sufficient to elucidate the
structure of all proteins of interest. While in early 2002
the protein sequence databases SWISS-PROT and
trEMBL contain details of over 250 000 proteins, only
about 10 000 of them have known 3D structures.
Furthermore, SWISS-PROT and trEMBL hold fewer
than 20 000 human proteins, meaning that many
others will be added in the near future. The direct
consequence of this is that only the ‘highly interesting’
proteins will be elucidated experimentally in the
foreseeable future.
In this context, comparative modeling methods
(homology based) have been developed and have
matured to a point where many of the resulting models
yield enough insights into a protein’s 3D structure to be
useful in functional analysis (Westhead and Thornton,
1998).
Advanced article
Article contents
 Introduction
 What is Comparative Protein Modeling?
 How Does One Build a Model?
 What Defines the Accuracy of a Model?
 About the Use of Protein Models
 Membrane Protein Models
doi: 10.1038/npg.els.0005273
Protein Homology Modeling
ENCYCLOPEDIA OF LIFE SCIENCES & 2005, John Wiley & Sons, Ltd. www.els.net 1
What is Comparative Protein
Modeling?
Proteins from different sources and with sometimes
diverse biological functions can have similar
sequences, and it is generally accepted that high
sequence similarity is reflected by distinct structural
similarity. Indeed, the relative mean square deviation
(rmsd) of the a-carbon coordinates for protein cores
sharing 50% residue identity is expected to be around
1 A˚ . Thus the most reliable prediction methods, called
comparative protein modeling (also often called
modeling by homology), consist of the extrapolation
of the structure for a new (or target) sequence from the
known 3D structure of related family members (or
templates). This process is guided by a sequence
alignment between target and template sequence (for
an overview of the modeling method, see Bajorath et al.
(1993)). Membrane proteins are almost completely
excluded from this approach, as there are only a very
few experimentally elucidated template structures
available (see the previous section). In this article we
will mainly focus on the accuracy and applicability of
protein models derived from these methods.
How Does One Build a Model?
Protein modeling is based on a combination of
methods used in bioinformatics and computational
chemistry, including searching sequence databases,
sequence threading, sequence alignment and force
field computations (energy minimization and molecular
dynamics). Four major steps can be distinguished
and these are broadly outlined below.
Identification of modeling templates
Comparative protein modeling requires at least one
sequence of known 3D structure with significant
similarity to the target sequence. In order to determine
if a modeling request can be carried out, one generally
compares the target sequence with a database of
sequences derived from the Brookhaven Protein
Data Bank (PDB). This can be performed using
sequence database search tools (FastA, BLAST
and PSI-BLAST, HMMER, PROFILE searches).
Generally speaking, the choice of template structures
should be restricted to those that share at least 25%
residue identity with 40% of the target sequence. These
limits are imposed by the current accuracy of the
modeling methods in general.
The above procedure might allow the selection of
several suitable templates, which can all be used but
must be optimally superposed in 3D. From this set of
superposed structures, a structurally corrected multiple
sequence alignment can then be derived.
Aligning the target sequence with the
template sequence(s)
The target sequence can then be aligned with the
template sequence or, if several templates are selected,
with the structurally corrected multiple sequence
alignment using the best-scoring diagonals obtained
by sequence alignment algorithms. Residues, which
should not be used for model building, for example
those located in nonconserved loops, should be
ignored during the modeling process. Thus, the
common core of the target protein and the loops
completely defined by at least one supplied template
structure will be built.
Building the model
Two very distinct classes of methods have been
developed to build models. One is based on the
satisfaction of spatial restraints derived from the
alignment between the target sequence and its 3D
templates. The other is based on an averaged framework
derived from the coordinates of the templates.
Rebuilding nonconserved loops can be performed
using a ‘spare-parts’ algorithm. Although most of the
known 3D structures available share no sequence or
structural similarity with the target and templates,
there might be similarities in the loop regions that
can be inserted into the protein model. Each loop is
defined by its length and the geometry of its ‘stems’,
namely the coordinates of the a-carbon (Ca) atom of
the four residues preceding and following the loop.
The fragments that correspond to the above loop
definition are extracted from the PDB entries if the
rmsd computed for their ‘stems’ is lower than a
specified cutoff value. Furthermore, only fragments
that do not overlap with neighboring parts of the
structure are considered possible candidates. The
accepted ‘spare parts’ are sorted according to their
rmsd and their degree of sequence similarity with the
target. The best-fitting fragment is then added to the
model.
Because the ‘spare-parts’ algorithm does not always
lead to convincing solutions, one can also use an
approach based on a conformational space search
driven by the satisfaction of stereochemical, distance
and steric constraints. Loops modeled with these
methods are filtered according to criteria such as the
surface exposure of hydrophobic moieties and relative
conformational energies.
The final step of coordinate generation is the
correction and completion of the side chains, using
Protein Homology Modeling
2
stereochemical criteria and libraries of allowed sidechain
conformations.
Model refinement
The final step of the model building process is the
idealization of the stereochemistry of the model, and
consists mainly of the optimization of bond geometry
and the removal of unfavorable nonbonded contacts.
This step is performed using energy minimization
methods as implemented in several force-field computation
packages. Excessive energy minimization will
cause the model to deviate markedly from the original
model, which is not suitable and should be avoided.
What Defines the Accuracy of
a Model?
The quality of a model is determined by two distinct
criteria, which will determine its applicability. First,
the correctness of a model is dictated by the quality of
the sequence alignment used to guide the modeling
process. If the sequence alignment is wrong in some
regions, then the spatial arrangement of the residues in
this portion of the model will be incorrect. The first
edition of the community-wide experiment known as
Critical Assessment of Protein Structure Prediction
(CASP) has already underscored that most severe
modeling errors can be traced back to mistakes in
sequence alignment (Mosimann et al., 1995). Despite
many efforts to address this issue (Jones and Kleywegt,
1999), it remains the main weakness of comparative
protein modeling. Second, the accuracy of a model is
essentially limited by the deviation of the template
structure(s) used relative to the experimental control
structure. This limitation is inherent to the methods
used, since models result from an extrapolation. As a
consequence, the core Ca atoms of protein models
which share 35À50% sequence identity with their
templates will generally deviate by 1.5À1.0 A˚ from
their experimental counterparts, as do experimentally
elucidated structures. One should, however, not
overlook the contributions of the templates to the
accuracy of the model. The templates, which are
obtained through experimental approaches, are subject
to structural variations caused not only by
experimental errors and differences in data collection
conditions À such as the temperature À but also by
different crystal lattice contacts and the presence or
absence of ligands. Furthermore, X-ray crystallography
and NMR generally yield 3D structures with an
even broader rmsd spread. This is well illustrated by a
typical example: the structure of interleukin 4 (IL-4)
(Harrison et al., 1995), a cytokine consisting of a
130-residue four-helix bundle, was elucidated by X-ray
crystallography as well as by NMR. The backbones of
three IL-4 crystal structures (PDB entries 1RCB, 2INT
and 1HIK) show rms deviations of 0.4À0.9 A˚ , while
those of three IL-4 NMR forms (PDB entries 1ITM,
1CYL and 2CYK) deviated by 1.2À2.6 A˚ . These
values illustrate the structural differences due to
experimental procedures and the molecular environment
at the time of data collection. It is thus crucial to
know the experimental conditions under which the
modeling templates were collected, as this has a direct
impact on the accuracy of the derived models and
thereby on their potential use.
Almost every protein model contains nonconserved
loops, which are expected to be the least reliable
portions of a protein model. Indeed, nonconserved
loops often deviate markedly from experimentally
determined control structures. In many cases, however,
these loops also correspond to the most flexible
parts of the structure, as evidenced by their high
crystallographic temperature factors (or multiple
solutions in NMR experiments). On the other hand,
the core residues À the least variable in any given
protein family À are usually found in essentially the
same orientation as in experimental control structures,
while far larger deviations are observed for surface
amino acids. This is expected since the core residues
are generally well conserved and the rotamers of their
side chains are constrained by neighboring residues. In
contrast, the more variable surface amino acids will
tend to show more deviations since there are few steric
constraints imposed upon them.
Some structural aspects of a protein model can be
verified using methods based on the inverse folding
approach. Two of them, namely the 3D profile-based
verification method and the Prosa suite of programs
developed by Manfred Sippl, are widely used. The 3D
profile of a protein structure is calculated by adding
the probability of occurrence for each residue in its 3D
context. Each of the 20 amino acids has a certain
probability of being located in one of the 18
environmental classes (defined by criteria such as the
amount of surface accessible to solvent, buried polar
and exposed nonpolar area, and secondary structure)
defined by Luthy et al. (1992). In contrast, ProsaII
relies on empirical pseudoconformational energy
potentials derived from the pairwise interactions
observed in well-defined protein structures. These
terms are summed over all residues in a model and
result in a more (more negative) or less (more positive)
favorable energy. Both methods can detect a global
sequence to structure incompatibility and errors
corresponding to topological differences between
template and target. They also allow the detection of
more localized errors such as b strands that are ‘out of
register’ or buried charged residues. These methods
Protein Homology Modeling
3
are, however, unable to detect the more subtle structural
inconsistencies often localized in nonconserved
loops, and cannot provide an assessment of the
correctness of their geometry.
About the Use of Protein Models
Protein models obtained using comparative modeling
methods can be classified into three broad categories:
 Models that are based on incorrect alignments
between target and template sequences. Such alignment
errors, which generally reside in the inaccurate
positioning of insertions and deletions, are caused
by the weaknesses of the alignment algorithms and
often cannot be resolved in the absence of a control
experimental structure. It is, however, often possible
to correct such errors by producing several models
based on alignment variants and by selecting the
most ‘sensible’ solution. Nevertheless, it turns out
that such models are often useful as the errors are
not located in the area of interest, such as within a
well-conserved active site.
 Models based on correct alignments are of course
much better, but their accuracy can still be medium
to low as the templates used during the modeling
process have a medium to low sequence similarity
with the target sequence. However, such models as
the ones described above are very useful tools for the
design of rational mutagenesis experiments. They
are not of great assistance during detailed ligandbinding
studies.
 The last category of models comprises all those
based on templates that share a high degree of
sequence identity (> 70%) with the target. Such
models have proved to be useful during drug design
projects and have allowed key decisions to be taken
in compound optimization and chemical synthesis.
For instance, models of several species variants of a
given enzyme can guide the design of more specific
nonnatural inhibitors.
However, nothing is absolute and there are numerous
occasions in which models falling into any of the
above categories could either not be used at all or, in
contrast, have proved to be more useful and correct
than initially thought. In our experience, several
applications of medium-accuracy models have proved
to be successful. These can be classified into three
categories detailed in the following subsections.
Interpreting the impact of mutations on
protein function: a potential link to diseases
One of the first uses that one can make of a model
structure is to interpret the impact a mutation can have
on the overall function of a protein. Although the
development of objective scoring functions has begun
only recently, ‘visual inspection’ associated with a
good knowledge of the rules underlying protein
structure has proved useful in defining the broad
reasons for mutant malfunction (Notarangelo et al.,
1996). When high-throughput production of singlenucleotide
polymorphisms (SNPs) is achieved, objective
scoring functions will be crucial in making
maximum use of the information. Indeed, a sizeable
proportion of the SNPs will alter the translated protein
sequences, and thus interpreting the potential functional
effects of these mutants will be crucial in
elucidating the molecular basis of human diseases.
Prioritization of residues to mutate to
determine protein function
As mentioned previously, the discovery of gene
function in the genomic era will require a sustained
experimental effort, which includes the creation of
molecular mutants. The prioritization of residues to
mutate will be greatly optimized by considering the 3D
structure of the target protein (Schneider et al., 1997).
Providing hints about protein function
This is probably the broadest and least well-defined
spectrum of potential applications for 3D models. The
common feature of these applications is that models
can be used to formulate a hypothesis around a
protein, which can then be tested in experimental
settings. It is well known that low, yet significant,
degrees of sequence similarity are often not sufficient
to attribute a function to a protein. In such cases,
protein modeling can provide useful insights and help
determine or confirm a potential functional assignment
(Duret et al., 1998). Furthermore, one can use
models to create a hypothesis about potential enzymatic
activities (Peitsch and Boguski, 1991) and
possible ligand-binding functions (Peitsch and
Boguski, 1990).
Membrane Protein Models
Membrane proteins remain a class of proteins
that represent an even greater challenge to modelers.
G-protein-coupled receptors, in particular, represent
a group of molecules of special interest to the
pharmaceutical industry, as a very large proportion
of today’s medicines are modulators of their activities.
Modeling such proteins has thus been attempted on
many occasions, and both de novo (Thomas, 1996) and
Protein Homology Modeling
4
comparative approaches (Thomas, 1996) have been
used. The two main steps along the path to a model
have been automated: algorithms have been developed
to identify the transmembrane domains (Persson et al.,
1996) and to generate 3D models using de novo
approaches (Herzyk and Hubbard, 1995) and comparative
methods. In all cases, however, the steps of
sequence analysis and coordinate generation were
separated and could not be linked automatically
because of the relatively low reliability of the first
step. Consequently this group of proteins is not yet
amenable to high-throughput model building. This
will of course dramatically change with the future
availability of an experimentally determined structure
for one of their family members.
References
Bajorath J, Stenkamp R and Aruffo A (1993) Knowledge-based
model building of proteins: concepts and examples. Protein
Science 2: 1798À1810.
Duret L, Guex N, Peitsch MC and Bairoch A (1998) New insulinlike
protein with atypical disulphide bond pattern characterised
Caenorhaditis elegans by comparative analysis and homology
modeling. Genome Research 8: 348À353.
Harrison RW, Chatterjee D and Weber IT (1995) Analysis of six
protein structures predicted by comparative modeling techniques.
Proteins: Structure, Function and Genetics 23: 463À471.
Herzyk P and Hubbard RE (1995) Automated method for modeling
seven-helix transmembrane receptors from experimental data.
Biophysical Journal 69: 2419À2442.
Jones TA and Kleywegt GJ (1999) CASP3 comparative modeling
evaluation. Proteins S3: 30À46.
Lu¨ thy R, Bowie JU and Eisenberg D (1992) Assessment of protein
models with three-dimensional profiles. Nature 356: 83À85.
Mosimann S, Melshko R and James MNG (1995) A critical
assessment of comparative modeling of tertiary structure of
proteins. Proteins 23: 301À317.
Notarangelo LD, Peitsch MC and Tore G Abrahamsen (1996)
CD40Lbase: a database of CD40L gene mutations causing Xlinked
hyper-IgM syndrome. Immunology Today 17: 511À516.
Peitsch MC and Boguski MS (1990) Is apolipoprotein D a
mammalian bilin-binding protein? New Biology 2: 197À206.
Peitsch MC and Boguski MS (1991) The first enzyme among the
lipocalin family. Trends in Biological Science 16: 363.
Persson B, Milpetz F and Argos P (1996) Prediction of transmembrane
segments in proteins using multiple sequence alignments.
In: Findlay JBC (ed.) Membrane Protein Models, pp. 1À25.
Oxford: BIOS Scientific.
Schneider P, Bodmer JL, Holler H, et al. (1997) Characterization of
the Fas (Apo-1, CD-95)ÀFas ligand (Apo-1L, CD95L) interaction.
Journal of Biological Chemistry 272: 18 827À18 833.
Thomas P (1996) Making and breaking models of G protein-coupled
receptors. In: Findlay JBC (ed.) Membrane Protein Models,
pp. 73À89. Oxford: Bios Scientific.
Westhead DR and Thornton JM (1998) Protein structure prediction.
Current Opinion in Biotechnology 9: 383À389.
Further Reading
Chothia C and Lesk AM (1986) The relation between the divergence
of sequence and structure in proteins. EMBO Journal 5: 823À826.
Herzyk P and Hubbard RE (1998) Using experimental information
to produce a model of the transmembrane domain of the ion
channel phospholamban. Biophysical Journal 74: 1203À1214.
Martin ACR, MacArthur MW and Thornton JM (1997) Assessment
of comparative modeling in CASP2. Proteins S1: 14À18.
Peitsch, MC (1997) Large scale protein modeling and model
repository. Proceedings of the Fifth International Conference on
Intelligent Systems for Molecular Biology, vol. 5, pp. 234À236.
Menlo Park, CA: AAAI Press.
Peitsch MC, Herzyk P, Wells TNC and Hubbard RE (1996)
Automated modeling of the transmembrane region of G-protein
coupled receptor by Swiss-Model. Receptors and Channels 4:
161À164.
Peitsch MC and Tschopp J (1995) Comparative molecular modeling
of the Fas-ligand and other members of the TNF family.
Molecular Immunology 32: 761À772.
Sankararamakrishnan R and Sansom MSP (1996) a-helix bundles
and ion channels. In: Findlay JBC (ed.) Membrane Protein
Models, pp. 55À72. Oxford: BIOS Scientific.
Sippl MJ (1993) Recognition of errors in three-dimensional
structures of proteins. Proteins: Structure, Function and Genetics
17: 355À362.
Tilton RF, Dewan JC and Petsko GA (1992) Effects of temperature
on protein structure and dynamics: X-ray crystallographic
studies of the protein ribonuclease-A at nine different temperatures
from 98 to 320 K. Biochemistry 31: 2469À2481.
von Heijne G (1992) Membrane protein structure prediction.
Hydrophobicity analysis and the positive-inside rule. Journal of
Molecular Biology 225: 487À494.
Protein Homology Modeling
5