Vol. 19 no.

18 2003, pages 2500–2501

BIOINFORMATICS APPLICATIONS NOTE DOI: 10.1093/bioinformatics/btg362

ModLoop: automated modeling of loops in

protein structures
András Fiser1, ∗ and Andrej Sali2,∗
1 Department of Biochemistry and Seaver Foundation Center for Bioinformatics,
Albert Einstein College of Medicine, 1300 Morris Park Ave., Bronx, NY 10461, USA
and 2 Departments of Biopharmaceutical Sciences and Pharmaceutical Chemistry,
California Institute for Quantitative Biomedical Research, Mission Bay Genentech
Hall, 600 16th Street, Suite N472D, University of California, San Francisco,
CA 94143-2240, USA

Received on May 1, 2003; revised on June 23, 2003; accepted on July 8, 2003

ABSTRACT (Fiser et al., 2002; Al Lazikani et al., 2001) in studying inter-

Summary: ModLoop is a web server for automated modeling actions between the protein and its ligands. Loop modeling
of loops in protein structures. The input is the atomic coordin- can also be useful in refining low- and medium-resolution
ates of the protein structure in the Protein Data Bank format, structures determined by X-ray crystallography and NMR
and the specification of the starting and ending residues of one spectroscopy.
or more segments to be modeled, containing no more than Prediction of loop conformations by optimization of an
20 residues in total. The output is the coordinates of the non- objective function was implemented in MODELLER (Fiser
hydrogen atoms in the modeled segments. A user provides et al., 2000). The method optimizes the positions of all
the input to the server via a simple web interface, and receives non-hydrogen atoms of a loop in a fixed environment. The
the output by e-mail. The server relies on the loop modeling optimization relies on a protocol consisting of the conjug-
routine in MODELLER that predicts the loop conformations by ate gradient minimization and molecular dynamics simulation
satisfaction of spatial restraints, without relying on a database with simulated annealing. The pseudo-energy function con-
of known protein structures. For a rapid response, ModLoop tains terms from a molecular mechanics force field as well as
runs on a cluster of Linux PC computers. restraints based on statistical distributions derived from known
Availability: The server is freely accessible to academic users protein structures. Bonds, angles, some dihedral angles and
at http://salilab.org/modloop improper dihedral angles are restrained by the correspond-
Contact: andras@fiserlab.org ing terms in the CHARMM-22 potential function (MacKerell
et al., 1998). The mainchain and sidechain dihedral angles
as well as non-bonded atom pairs are restrained by statistical
The function of a protein is generally determined by
potentials extracted from many known protein structures (Sali
shape, dynamics and physiochemical properties of its solvent
and Blundell, 1993; Sali and Overington, 1994; Fiser et al.,
exposed molecular surface. Likewise, functional differences
2000; Melo and Feytmans, 1997).
among the members of the same protein family are usually a
Evaluation of the method relied on 40 randomly selected
consequence of the structural differences on the protein sur-
loops of known structure at each length from 1 to 14 residues.
face. In a given fold family, structural variability is a result
The accuracy was determined by building loops both in the
of substitutions, insertions and deletions of residues among
native and distorted loop environments because only approx-
members of the family. Such changes frequently correspond
imate loop environments are available in real comparative
to exposed loop regions that connect elements of secondary
modeling applications. The errors in loop predictions increase
structure in the protein fold. Thus, loops often contribute to
with loop length and environment distortion (Fig. 1).
binding sites and determine the functional specificity of a
The modeling protocol generates a number of independ-
given protein framework. Consequently, the accuracy of loop
ently optimized conformations, starting with random initial
modeling (Oliva et al., 1997; Fine et al., 1986; Xiang et al.,
conformations. The final loop prediction is the optimized con-
2002; van Vlijmen and Karplus, 1997; Rapp and Friesner,
formation that has the lowest pseudo-energy score. Currently,
1999; Martin and Thornton, 1996) is a major factor determ-
the number of independent optimizations is limited to 300.
ining the usefulness of comparative protein structure models
Because each individual optimization typically takes a few
minutes of CPU time on a Pentium processor, the ModLoop
∗ To whom correspondence should be addressed. web server runs on a cluster of Pentium nodes.

2500 Bioinformatics 19(18) © Oxford University Press 2003; all rights reserved.
 Automated modeling of loops in protein structures

Barrientos,L.G., Campos-Olivas,R., Louis,J.M., Fiser,A., Sali,A.

and Gronenborn,A.M. (2001) 1H, 13C, 15N resonance assign-
ments and fold verification of a circular permuted variant of the
potent HIV-inactivating protein cyanovirin-N. J. Biomol. NMR,
19, 289–290.
Fine,R.M., Wang,H., Shenkin,P.S., Yarmush,D.L. and Levinthal,C.
(1986) Predicting antibody hypervariable loop conformations.
II: minimization and molecular dynamics studies of MCPC603
from many randomly generated loop conformations. Proteins, 1,
342–362.
Fiser,A., Do,R.K. and Sali,A. (2000) Modeling of loops in protein
structures. Protein Sci., 9, 1753–1773.
Fiser,A., Feig,M., Brooks,C.L.,III. and Sali,A. (2002) Evolution and
Fig. 1. The accuracy of loop modeling as a function of environment physics in comparative protein structure modeling. Acc. Chem.
distortion (Fiser et al., 2000). Average prediction accuracy is shown Res., 35, 413–421.
for 40 loops of 4 (open circles), 8 (filled circles) and 12 residues Lee,S.A., Shen,E.L., Fiser,A., Sali,A. and Guo,S. (2003)
(filled squares). The root-mean-square (RMS) error of the loops is The zebrafish forkhead transcription factor Foxi1 specifies
calculated for the four mainchain atom types upon the optimal super- epibranchial placode-derived sensory neurons. Development, 130,
position of the whole model on the native structure. The RMS error 2669–2679.
upon superposition of only the loop atoms tends to be ∼1.5 times MacKerell,A.D.,Jr, Bashford,D., Bellott,M., Dunbrack,R.L.,Jr,
smaller for 8-residue loops than the errors shown (Fiser et al., 2000). Evanseck,J.D., Field,M.J., Fischer,S., Gao,J., Guo,H., Ha,S.
The distortion of the environment is measured by the RMS devi- et al. (1998) All-atom empirical potential for molecular mod-
ation for the three residues flanking both sides of the loop. Error bars leing and dynamics studies of proteins. J. Phys. Chem. B, 102,
indicate the standard error of the mean. 3586–3616.
Martin,A.C. and Thornton,J.M. (1996) Structural families in loops
of homologous proteins: automatic classification, modelling and
application to antibodies. J. Mol. Biol., 263, 800–815.
The ModLoop server requires as input a coordinate file in Melo,F. and Feytmans,E. (1997) Novel knowledge-based mean force
the Protein Data Bank (PDB) format, as well as the starting potential at atomic level. J. Mol. Biol., 267, 207–222.
and ending positions of the loops. The user can specify several Oliva,B., Bates,P.A., Querol,E., Aviles,F.X. and Sternberg,M.J.
loops as input. These loops will be optimized simultaneously, (1997) An automated classification of the structure of protein
which is particularly useful if conformations of multiple inter- loops. J. Mol. Biol., 266, 814–830.
acting loops need to be predicted. Given the rapid decrease in Rapp,C.S. and Friesner,R.A. (1999) Prediction of loop geometries
the prediction accuracy as the loop length increases, the total using a generalized born model of solvation effects. Proteins, 35,
173–183.
number of residues in all selected loops is currently limited
Sali,A. and Blundell,T.L. (1993) Comparative protein model-
to 20. The resulting coordinate file in the PDB format is sent ling by satisfaction of spatial restraints. J. Mol. Biol., 234,
back to the user by e-mail. 779–815.
The ModLoop server is a useful addition to the already avail- Sali,A. and Overington,J.P. (1994) Derivation of rules for com-
able protein structure modeling servers (http://salilab.org/ parative protein modeling from a database of protein structure
bioinformatics_resources.shtml). It has been utilized in sev- alignments. Protein Sci., 3, 1582–1596.
eral applications, including modeling of the active site loop van Vlijmen,H.W. and Karplus,M. (1997) PDB-based protein loop
conformations in dehydrogenases with various substrate spe- prediction: parameters for selection and methods for optimization.
cificities (Wu et al., 1999; Vernal et al., 2002), predicting J. Mol. Biol., 267, 975–1001.
the conformation of the linker loop in an artificial con- Vernal,J., Fiser,A., Sali,A., Muller,M., Jose,C.J. and Nowicki,C.
struct of a circularly permuted cyanovirin protein (Barrientos (2002) Probing the specificity of a trypanosomal aromatic alpha-
hydroxy acid dehydrogenase by site-directed mutagenesis. Bio-
et al., 2001), and rationalizing the observed functional impact
chem. Biophys. Res. Commun., 293, 633–639.
of various mutants of the zebrafish winged helix protein Wu,G., Fiser,A., ter Kuile,B., Sali,A. and Muller,M. (1999) Conver-
No Soul/Foxi1 (Lee et al., 2003). gent evolution of Trichomonas vaginalis lactate dehydrogenase
from malate dehydrogenase. Proc. Natl Acad. Sci. USA, 96,
6285–6290.
REFERENCES Xiang,Z., Soto,C.S. and Honig,B. (2002) Evaluating conformational
Al Lazikani,B., Jung,J., Xiang,Z. and Honig,B. (2001) Protein free energies: the colony energy and its application to the problem
structure prediction. Curr. Opin. Chem. Biol., 5, 51–56. of loop prediction. Proc. Natl Acad. Sci. USA, 99, 7432–7437.

2501