- BK21, Division of Chemistry and Molecular Engineering, SNU, KRF
- Basic Research Program, KOSEF
- Marine Biotechnology Program funded by Ministry of Land, Transport and Maritime Affairs
- The ultimate goal of our research is to understand and to predict the functions of biomolecules in the atomistic details with theoretical and computational methods
and to use the knowledge to contribute to design of molecules with novel functions, such as new drugs, new catalysts, or new materials.
For such a purpose, we have been studying various areas of biomolecular structure and interaction. The main theme of our current approach is
to accurately compute changes in energy and structure accompanied by biomolecular interactions that are relevant to physiological functions.
Various research techniques are needed for such studies. We have been developing some of the techniques ourselves, for example,
protein loop modeling and protein-ligand docking techniques, and we are also using available tools at the same time when well-established tools,
such as molecular dynamics simulation packages, are available.
The research topics that we have been investigating can be divided into the following three major categories:
1) algorithm development for computational prediction of protein structures,
2) molecular dynamics simulations of conformational changes of biomolecules, and
3) algorithm development for prediction of protein-ligand interactions.
Computational prediction of protein structures forms a basis of many computational studies when experimental structures are not available
due to difficulties in experiments, for example in the cases of transmembrane proteins.
Computational study of conformational changes of biomolecules provides deeper understanding on biological function in the atomic details
and knowledge on essential factors that influence physiological functions that are necessary for designing molecules that regulate the functions.
Computational prediction of protein-ligand interaction provides a tool for prediction of biological functions and for design of small molecules
that inhibit disease-related processes. The three different topics are connected with each other, for example,
better structure prediction method can be combined with better protein-ligand docking method to understand a biological process better,
and a better understanding of conformational changes that accompany binding of a ligand to a receptor can be utilized to design a better antagonist or agonist
for the process of interest. Therefore, the current efforts in different areas of computational approaches will eventually converge
towards the goal of understanding and regulation of biomolecular processes in the future.
Protein Structure Prediction
Predicting protein structure from the amino acid sequence is a fundamental problem in protein science. Our approach is based on the analytical loop closure algorithm [1, 2],
which is the most general loop closure algorithm developed for protein loop modeling so far. After the first paper , we refined the mathematical aspects further,
and made the loop closure algorithm more efficient with the use of the Dixon resultant instead of the Sylvester resultant . The loop closure algorithm was also applied
to a small ring system of proline, and correlation of the ring puckering and the backbone conformation was explained with the analytical method .
A more interesting application is to modeling of longer protein loops because accurate prediction of protein loop structures
is a main obstacle in accurate protein structure prediction. Protein loops have irregular structures that are determined by the amino acid sequence and the surrounding protein environment.
Structures of protein loops are harder to predict accurately compared to those of the regions with templates for which informatics-based method can be applied.
There are no proper templates for protein loops, and their structures must be computed with an ab initio method.
Moreover, predicting longer loops is even more challenging when physics-based method is employed because of the larger degrees of freedom.
To overcome this difficulty, we are currently employing a combination of physics-based method and informatics-based method.
In the combined method, fragment structures are collected from the existing structure database according to the 'local' sequence similarities,
and they are assembled into loop structures. During the course of fragment assembly and refinement, physical interactions of the loop and the environment are
fully taken into account. Only a preliminary study has been published so far , but more extensive studies are on-going to prove the applicability of the method to a large set of protein loops.
A successful extension in this direction will result in a very useful protein modeling tool for the community. Immediate application is of course to modeling of the loop regions
when template-based method is used, and other applications include prediction of flexible loop structures upon protein-ligand or protein-protein binding.
One interesting theoretical development with the loop closure algorithm is the derivation of the analytical gradient (paper in preparation).
The analytical gradient can be utilized when an objective function of coordinates under the loop closure constraint is to be optimized.
An example of such an objective function is a biasing potential that favors particular physical interactions, e.g. known hydrogen bonds, in a receptor-ligand complex.
Such a biasing potential can improve the computational efficiency of the conformational search process greatly by directly guiding to the desired interactions that are known empirically or intuitively.
We plan to apply the method to impose flexibility efficiently in computational prediction of bound structures of protein and ligand or protein and protein.
Finally, for accurate and efficient modeling of protein structures we have developed our own molecular mechanics algorithms using quaternions [5, 6, 7].
An efficient algorithm for geometry manipulations (conversion of internal-to-Cartesian coordinates) of chain molecules has been developed using quaternions ,
although it turned out to be equivalent to the method developed by Thompson in 1967 (J. Chem. Phys).
We also derived an algorithm of calculating best-fit RMSD of two structures using quaternions .
This method is now being introduced as a representative method for calculation of RMSD in the field together with the old method of Kabsch (1976, Acta Crystallographica).
[Entry in Wikipedia]
A proof of the equivalence of the two methods was also given in Ref. 7.
All those algorithms, including the analytical loop closure and its gradient are incorporated in our own molecular mechanics program, SNULOOP.
Collaborations with the following groups are in progress: Prof. Jooyoung Lee (School of Computational Sciences, KIAS) for application of the loop closure algorithm to homology modeling,
Prof. Julian Lee (Department of Bioinformatics and Life Science, Soongsil Univeristy) for loop modeling with fragment assembly, and
Prof. Evangelos Coutsias (Department of Mathematics and Statistics, University of New Mexico) for analytical gradient of loop closure.
Conformational Changes of Biomolecules
In addition to the method development efforts related to computation of protein structures, we are also using available program packages (AMBER, CHARMM, and GROMACS)
to study conformational changes of biomolecules with molecular dynamics simulations. Large conformational changes are not easy to observe in naive molecular dynamics
simulation runs, so techniques such as generalized ensemble simulations, umbrella sampling, targeted molecular dynamics simulations, replica path simulations, etc
must be employed.
We have developed an analysis method to be used to obtain ensemble averages and potential of mean force from generalized ensemble simulations or
umbrella sampling . The analysis method was applied to the conformations of melittin ions probed with the Electron Capture Dissociation Mass Spectrometry
(work with Prof. Hahnbin Oh in Chemistry, Sogang U) (to be submitted). A molecular dynamics simulation study was also performed to investigate the mechanism
behind the functional conversion by a single amino acid in flowering regulator proteins .
One interesting problem we are studying is conformational changes in DNA that involve Z-DNA. Z-DNA is the first single crystal structure discovered in 1979 and is left-handed
unlike the righ-handed B-DNA structure that was proposed by Watson and Crick in 1953. Z-DNA is stabilized by negative supercoiling and is thought to play a role in transcription.
We performed molecular dynamics simulations (targeted molecular dynamics simulations and umbrella sampling using AMBER)
to find possible transition pathway from the B-DNA to the Z-DNA structure (paper in preparation). Our approach is very different from previous simulations
in that we consider the B-Z junction structure and simulate a movement of the junction. This pathway gave much lower free energy barrier than before,
implying that such a transition is much more feasible than previously proposed mechanism. With Dr. Bernard Brooks at NIH, we are now applying a replica path method
to investigate other possible pathways. Study on effects of negative supercoiling on DNA structure is also underway (collaboration with Prof. S. Hong, Department of Physics, Korea University).
Another interesting system we are working on is the arspartate chemotaxis receptor. The structures above and below the transmembrane regions of the receptor are known,
and only the transmembrane regions are unknown. There are hypotheses on the mechanism of signal transduction upon ligand binding, but there is no definite answer
due to the absence of structural information on the transmembrane region. This problem is hard to be attacked with experimental techniques because of the presence
of the lipid bilayer, but it is a perfect problem for a computational biologist. We are now performing a large-scale simulation using the GROMACS program.
(in collaboration with Prof. Wonpil Im, Kansas University)
Computational study of transmembrane proteins is a fascinating area, and the first step would be to develop a proper energy function for transmembrane proteins.
In the course of the research, we became interested in the C-alpha hydrogen bonds that are observed frequently in the high-resolution crystal structures
of transmembrane proteins. Two JACS papers argued that the hydrogen bond is not stable at all or only marginally stable from mutation and FTIR studies,
although ab initio quantum mechanical studies showed that such hydrogen bonds are about half as stable as conventional hydrogen bonds.
We calculated the strengths of the C-alpha hydrogen bonds in the actual crystal structures of transmembrane proteins with quantum mechanical methods using GAUSSIAN,
and indentified the cases when the hydrogen bonds are stable or unstable .
These studies are just a beginning of the long-term goal of prediction and understanding of transmembrane protein structures and their roles in biological function.
Many protein-ligand docking algorithms are available as commercial and academic programs, and they are being used actively for lead discovery and lead optimization in numerous labs
in the world. Fast docking programs are conveniently used for virtual screenings of compound databases. More expensive computations can also be used to predict docking
poses and binding affinities more accurately for lead optimization.
We are currently developing our own docking program with a better performance (funded by KOSEF). New energy functions and new algorithms are also being developed.
Two major elements needed for a docking program are a conformational search algorithm and a scoring function (energy function). We are currently developing both.
We have recently developed a rescoring function  that improves prediction of binding poses of protein-ligand complexes.
This was possible by including entropic effect with a concept of colony energy to the current scoring functions that ignore such effect.
The rescoring function was developed such that scores of conformations generated from available docking programs are re-evaluated to improve prediction of docking poses.
Tests on 11 different scoring functions were successfully performed. The next step is to extend the rescoring method to develop a better scoring function
that accurately predicts binding affinities as well as binding poses. Such a study is underway. Solvation models are also being investigated to improve the prediction accuracy.
Our approach to development of a conformational sampling method is again
based on the analytical loop closure algorithm. The loop closure algorithm is being applied to sample flexible receptor conformations
because an efficient local backbone perturbation is possible with the algorithm. One of the hot issues in the protein-ligand docking community is how
to incorporate receptor flexibility. A lot of efforts are being made to develop a better method, but the major problem is the computational efficiency.
Consideration of fully flexible receptor such as in molecular dynamics simulations is impractical because of the huge computational cost.
Side chain flexibilities near the binding site at fixed backbone structures are incorporated in some docking packages, but it is not seem to be enough.
Some methods take several experimental structures bound to different ligands or structures generated from computational methods such as molecular dynamics simulation
or network calculation. Docking on such individual proteins or combined structures are performed, but the results are not satisfactory.
Our strategy is very unique in that backbone flexibilities as well as side chain flexibilities are directly allowed in the binding pocket,
and the flexibilities in the presence of a ligand of interest are realistically considered.
This will be possible because of the efficiency of the analytical loop closure algorithm.
Because our own docking program is still under development, we are currently using available docking programs such as DOCK, AutoDock, and FlexX for practical application
to actual biological systems. Collaborations with Prof. Sangho Choi (Food Science and Technology, SNU) for discovery of ligands of proteins
involved in the quorum sensing pathway in bacteria (LuxS, LuxP, and SmcR) and with Heonjoong Kang (Earth and Environmental Sciences, SNU)
for discovery of agonists of PPAR-d and FXR are in progress.
Collaborations with experimental groups (structural biology, plant biology, biotechnology, etc) are underway.
Various systems such as protein loops, protein-ligand complexes, DNA molecules, and membrane proteins are being studied with various techniques
such as the analytical loop closure, protein-ligand docking, and molecular dynamics simulations. I believe that available methodologies must be utilized
to study current problems of interests, and efforts in method development must also be continued at the same time to expand our capacity further.
Our group is working in both directions, and such combined efforts will contribute to understanding, prediction, and regulation of biomolecular systems
to produce fruitful results.
 E. A. Coutsias, C. Seok*, M. P. Jacobson, and K. A. Dill, "A Kinematic View of Loop Closure", J. Comput. Chem. 25, 510 (2004).
 E. A. Coutsias, C. Seok, M. J. Wester, and K. A. Dill, "Resultants and Loop Closure", Int. J. Quantum Chem. 106, 176 (2006).
 B. K. Ho, E. A. Coutsias, C. Seok, and K. A. Dill, "The flexibility in the proline ring couples to the protein backbone", Protein Sci, 14, 1011 (2005).
 D. Lee, C. Seok*, and J. Lee*, Protein loop modeling using fragment assembly, J. Korean Phys. Soc. 52, 1137 (2008).
 C. Seok* and E. A. Coutsias*, Efficiency of rotational operators for geometric manipulation of chain molecules, Bull. Korean Chem. Soc. 28, 1705 (2007).
 E. A. Coutsias, C. Seok*, and K. A. Dill, "Using Quaternions to Calculate RMSD", J. Comput. Chem. 25, 1849 (2004).
 E. A. Coutsias, C. Seok*, and K. A. Dill, "Rotational Superposition and least squares: The SVD and Quaternions approaches yield identical results", J. Comput. Chem. 26, 1663 (2005).
 J. Lee and C. Seok*, A statistical rescoring scheme for protein-ligand docking: Consideration of entropic effect, Proteins: Structure, Function, and Bioinformatics, 70, 1074-1083 (2008).
 J. D. Chodera, W. C. Swope, J. W. Pitera, C. Seok, and K. A. Dill, "Use of the Weighted Histogram Analysis Method for the Analysis of Simulated and Parallel Tempering Simulations", Journal of Chemical Theory and Computation, 3, 26 (2007).
 S. Kang, J. Lee, M. S. Lee, and C. Seok*, Structural basis of functional conversion of a floral repressor to an activator, Bull. Korean Chem. Soc. 29, 408 (2008).
 H. Park, J. Yoon, and C. Seok*, Strength of Ca-H...O=C Hydrogen Bonds in Transmembrane Proteins, J. Phys. Chem. B, 112, 1041-1048 (2008).