University of Southern California Dissertations and Theses

Computational modeling of solvation and docking of peptide-MHC class I

(USC Thesis Other)

Computational modeling of solvation and docking of peptide-MHC class I

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1

Computational Modeling of Solvation and Docking of Peptide-MHC Class I

Jiawei Lai

PHARMACEUTICAL SCIENCES
MASTER OF SCIENCE
University of Southern California
School of Pharmacy

August 2019

2

Table of Contents
Abstract 3
Chapter 1: Background 4
1.1 Peptide-MHC complexes 4
1.2 MHC Class I docking software 5
Chapter 2: Solvation of peptide in protein: WATGEN 7
2.1 Introduction 7
2.2 Methods 8
2.3 Results 10
2.3.1 X-ray crystallography vs. WATGEN predicted water networks 10
2.3.2 WATGEN predicted water networks before and after peptide binding to MHC 18
2.3.3 Displaced water and its potential in docking 22
2.4 Discussion 23
Chapter 3: Docking peptide in MHC class I molecules: EXSAN 24
3.1 Introduction 24
3.2 Methods 25
3.3 Results and discussion 29
3.3.1 Validation of EXSAN 29
3.3.2 X-ray structure vs. EXSAN predicted peptide conformation 36
Chapter 4: Evaluation and Conclusion 50
4.1 Further study and Improvements 50
4.2 Conclusion 51
Supplement 52
References 54

3

Abstract
This thesis demonstrates the potential of computational methods to predict the peptide
structure in peptide-MHC class I complexes. There are two parts of the thesis: 1. modeling the
water network in the peptide-MHC class I complexes using WATGEN; 2. incorporating water
networks in predicting the peptide conformation using EXSAN. The first part demonstrates the
capability of obtaining a reliable water network using only a target protein and peptide. The
second part shows the potential of predicting bound peptide conformation in the peptide-MHC
class I complex. The focus is to discuss the potential of these two programs and how they
integrate for predicting peptide-protein complexes.

4

Chapter 1: Background
1.1 Peptide-MHC complexes
Peptide- major histocompatibility complex (MHC) complex has been extensively studied
in the past decades. In the Protein Data Bank, there are abundant sources of X-ray
crystallography structures for viewing. Even with enough information, the complexities and
variability of MHC molecules are still an important area for further exploration. This research is
based on the known X-ray structure of MHC molecules, and we would like to predict the
presence of water in the binding groove of the peptide to the MHC. By predicting the position of
water molecules inside the binding groove, we can have more insight into the interactions
between the peptide and protein. We also used a computational method to dock peptide to MHC
molecules, offering a practical solution for testing potential peptides that fit the MHC.
MHC class I is present in all cells other than red blood cells. It takes intracellular antigens
that are degraded by proteases in cytosol and enter the endoplasmic reticulum. It then transports
the antigens to the cell membrane and presents the foreign antigens to cytotoxic T cells, which
can recognize the viral peptides and kills infected cells. Types of peptides accepted by the MHC
class I: typically, MHC class I accepts eight to ten amino acids peptide, but in some cases, MHC
class I can also accept up to 14 amino acids peptides - in these special cases, the long-budged
epitope can’t be effectively recognized by the T cell receptor. In some recent studies, it is shown
that these long-bulged peptides can also be recognized by the TCR.

5

1.2 MHC Class I docking software
There are several groups that also developed MHC class I docking software, such as
DINC, GradDock, and DynaPred.
DINC (Antunes et al., 2018) uses AutoDock4, which uses Lamarchian genetic algorithm,
a type of evolutionary technique that is commonly used for stochastic sampling of ligand
conformations in molecular docking. The scoring uses a semi-empirical free energy force field.
DINC is an incremental meta-docking approach that can predict the binding modes of large
peptide ligands bound to MHC receptors. It is useful in allowing docking for different MHC
allotypes, predicting unusual binding modes, and obtaining accurate structural prediction for
peptides with up to 41 rotatable bonds. DINC has had a positive impact in many fields related to
human health, including vaccine development and tissue transplantation. Fast and accurate
prediction of peptide-specific pMHC complexes will be key for development of safe and
effective T-cell-based immunotherapies against cancer.
GradDock (Kyeong et al., 2018) is a MHC class I docking software that uses Monte
Carlo method algorithm. It has three major steps: generate peptide candidates from the sequence;
insertion of peptides into MHC class I molecules; ranking of peptides using Rosetta score terms
and then normalizing the score. It assumes that the crystal structure is in the minimum energy
state, so each peptide constitutes energy inequalities against its native crystal structure.
GradDock’s successful simulation must satisfy these two criteria: near-native decoys must exist
and a wide range of competitive decoys must be included.
DynaPred (Antes et al., 2006) is a structure and sequence based method for MHC class I
binding peptide that can produce docking results in a simple and efficient way. There are two
major steps for the prediction, prediction of the binders from peptide sequence database, and then
6

the construction of peptide-MHC class I complexes using the sequence database. There are two
prediction models including the position-dependent method and position independent method.
The position-dependent method is capable of achieving an average backbone RMSD of 1.53 Å
compared to the experimental results. The general approach of DynaPred is to dock a peptide to
a MHC class I molecule based on an approximation of binding free energy of amino acids to the
binding pockets in MHC class I. The assumption of DynaPred’s algorithm is that the affinity of
the individual amino acid should sum up to be similar to the affinity of the whole peptide. It is
clear that it has a totally different approach compare to EXSAN, which considers the docking
energy of the whole peptide instead of the individual amino acid residue. The advantage of
calculating the energy by individual amino acid residue is to have a quick assessment of how
well a particular amino acid fits inside the binding pocket, instead of calculating how the
inclusion of the amino acid affects the overall conformation of the peptide in the binding pocket.
As a result, DynaPred is capable of predicting peptide-MHC complexes rapidly to examine
whether a peptide sequence from the database is a binder, and if a sequence is classified as
binder, DynaPred can construct the bound conformation of the potential binder. The limitation of
DynaPred is that it was only tested on HLA-A2.

7

Chapter 2: Solvation of peptide-protein complexes: WATGEN
2.1 Introduction
Water is an important part of a peptide-protein binding interface. By evaluating and
establishing a water network, we can have better understanding of how the peptide binds to the
target protein. WATGEN is an algorithm for modeling water networks at protein-protein
interfaces (Bui et al., 2007). The goal for WATGEN is to predict the water around the peptide
and protein and predict how the water plays a role in the binding. The relationships include water
to protein, water to peptide, and water to water interactions. Some water molecules are able to
form water-mediated salt bridges to connect a peptide to the target protein. WATGEN can also
be used for studying how the peptide displaces water molecules in the binding pocket, which is
useful for understanding why certain conformations of the peptide are more favorable than others
and why certain peptides have higher affinity to the target protein.
Calculation of the interfacial water network is used for protein-peptide complexes or
interfaces. The input files to WATGEN include X-ray structures of the protein and peptide in
PDB files. The user can determine the receptor and ligand in WATGEN. WATGEN uses four
steps to predict the water structure between peptide and protein. 1) distribution of water
molecules, 2) scoring of water molecules according to the interactions with neighboring protein
or water molecules, 3) selection of the best water molecules, and 4) optimization of the water
network. Oxygen atoms are added, followed by addition of hydrogen atoms (Bui et al., 2007).
WATGEN can predict the positions of water molecules in a logical way, and the
construction of the water network is critical in predicting the binding of peptides to proteins. Due
to the limit in resolution of X-ray crystallography, the full water network in the X-ray structure
could be missing, and so a computationally generated water network can help optimize the
8

protein-peptide binding.

2.2 Methods
WATGEN was used to predict water networks in multiple peptide-MHC class I
complexes with X-ray structures. The water structure was also compared between the MHC class
I complex and MHC class I protein only, and the relationship between affinity and the amount of
displaced water molecules was examined. X-ray structures of MHC class I molecules were
retrieved from Protein Data Bank.
Depending on the location and interactions of the water molecules, there are several
different interactions made by water molecules, such as hydrogen bonds, single water bridges,
double water bridges, and buried water. Hydrogen bonding water molecules form hydrogen
bonds with other water molecules, peptide, or protein. For a hydrophilic residue on the peptide, it
is favorable to form a hydrogen bond with water molecules in the binding site or a hydrophilic
residue on the protein. To make such a comparison, we examined the water network inside the
cavity before and after binding of the peptide to the binding site. A single water bridge is
assigned to water molecules that form a water bridge with other hydrophilic residue on the
protein and also form a water bridge with a hydrophilic part of the peptide. A double water
bridge involves two water molecules that form a hydrogen bond with each other, and one of
these two water molecules forms a water bridge with hydrophilic residue on the protein and the
other forms a water bridge with the peptide. Buried water are molecules that fail to find
interactions with other water molecules and link all the way to the bulk solution. This type of
water molecule can be found mainly in hydrophobic pocket in the binding site. It is relatively
less favorable to place a hydrophilic component in the hydrophobic pocket, and having such
9

water molecules in the hydrophobic pocket is not desirable.
Since we assigned different scores for different water types, we can then evaluate the
water network by considering the types of water molecules. It also allows a comparison of the
differences between an empty MHC class I and the peptide-MHC class I complex. Upon the
binding of peptides to MHC class I, water is displaced by the peptide, and we can use the number
and types of displaced water to estimate the binding affinity. A total of 58 different peptide-
MHC class I complexes were retrieved from the Protein Data Bank. Each of the complexes was
run in WATGEN twice: a run on the peptide-MHC class I complex and a run on the MHC class I
alone. The predicted water networks from the two runs are compared and contrasted to provide
information on displaced water molecules that can help improve docking.

10

2.3 Results
2.3.1. X-ray crystallography vs. WATGEN predicted water networks
The water structure of NY-ESO-1 epitope, SLLMWITQC, in complex with HLA-A2
heavy chain is shown in Protein Data Bank (PDB) Code: 1S9W (Webb et al., 2004). This
structure was used to test the ability of WATGEN to predict water, particularly water molecules
in key positions in the complex. In this run, WATGEN located water molecules that play
important roles in peptide binding. In Fig 2.3.1.B, the green dots are oxygen molecules that show
the location of WATGEN-predicted water molecules. Water molecules around the peptide and
the binding site play important roles in determining the affinity of the binding.

Figure 2.3.1A Comparison of X-ray water molecules (red) and WATGEN-predicted water
molecules (green) while the peptide is in the cleft of HLA-A2.
11

Figure 2.3.1B Comparison of X-ray water molecules (red) and WATGEN predicted water
molecules (green) while the peptide is in the cleft of HLA-A2.

Figure 2.3.1A & B show the output of the WATGEN-predicted water network and a
comparison between WATGEN water molecules and X-ray detected water molecules. In this
run, water molecules between peptide and MHC class I are considered and predicted because
they are significant in the affinity of the peptide binding to the MHC cleft. The following figures
show a closer look at predicted water molecules and how well they represent the actual positions
of X-ray water.
12

Figure 2.3.1C Comparison of X-ray water (Red) and WATGEN-predicted water (Green). Water
molecules mediate interaction between Cys 9 on the peptide and Thr 80 on MHC class I.

Figure 2.3.1C is a closer look at the C-terminal cysteine and its interaction with water
molecules and HLA-A2. From X-ray crystallography, the nitrogen on the Cys 9 backbone is 3.26
Å away from water HOH327, and then HOH327 is 2.72 Å away from the OG1 of Thr 80 on the
HLA-A2. WATGEN predicts a water molecule WAT3 only 0.76 Å away from the correct X-ray
water. This water molecule is a key component in the water-mediated hydrogen bond between
peptide and HLA-A2.
13

Figure 2.3.1D Comparison of X-ray water (Red) and WATGEN-predicted water (Green).
Water molecules mediate the interaction between Cys 9 on the peptide and Asp 77 on MHC
class I (other water molecules were removed for a clean view of the structure).

Similarly, from Figure 2.3.1D, X-way detected HOH327 also facilitates a water mediated
interaction between the terminal oxygen on the Cys 9 and OG1 of Thr 80 in HLA-A2 and OD1
of Asp 77 in HLA-A2. WAT3 is has a similar role to that of HOH327.
14

Figure 2.3.1E Comparison of X-ray water (Red) and WATGEN-predicted water (Green). Water
molecules mediate an interaction between Gln 8 on the peptide and Thr 73 on MHC class I.

From Figure 2.3.1E, HOH318 forms a water-mediated bond between the NE2 of Gln 8
and OG1 from Thr 73. WATGEN correctly predicts the location of this water molecule with a
minor difference of 0.22 Å compare to the X-ray detected water.
15

Figure 2.3.1F Comparison of X-ray water (Red) and WATGEN-predicted water (Green). Water
molecules mediate an interaction between Thr7 on the peptide and Gln 155 on MHC class I.

WATGEN is not perfect at getting all the key water molecules correct. In Figure 2.3.1F,
an X-ray-detected water molecule HOH312 mediates an interaction between OG1 from Thr 7 of
the peptide and OE1 from Gln 155 form the HLA. WATGEN failed to predict this water
molecule, and the closest water, WAT4, is 1.74 Å away from the X-ray water.
16

Figure 2.3.1G Comparison of X-ray water (Red) and WATGEN-predicted water (Green).
Water molecules mediate interactions between Thr 7 on the peptide and Arg 97 on MHC class I.

Figure 2.3.1G shows a key interaction between the positively charged Arg 77 on the beta
sheet of the HLA-A cleft and the oxygen of the peptide backbone. The guanidinium group of the
arginine can interact with two water molecules in the cleft, and then these water molecules form
water-mediated interactions with the oxygen atom on the backbone. The distance between X-ray
water HOH110 to NH1 on Arg 97 is 3.18 Å; the distance between HOH110 to NH2 on Arg 97 is
2.67 Å. WATGEN-predicted WAT17 is about 3 Å away from the side chain of Arg 97.
17

WATGEN is close enough to predict only one atom in the middle of these two detected
water molecules (HOH110 and HOH289). This WATGEN predicted water (WAT17) resembles
the X-ray detected water in a reasonable position. Ideally, there should be two molecules
predicted just as the X-ray structure, but this is the best WATGEN can achieve at present.

Figure 2.3.1G Comparison of X-ray water (Red) and WATGEN-predicted water (Green). Water
molecules mediate an interaction between Ile 6 on the peptide and Ala 69 on MHC class I (other
water molecules, both X-ray and predicted, are removed for simplicity of this figure).

In Figure 2.3.1G, the nitrogen atom on Ile6 interact with X-ray HOH 48 and HOH 48
then also interacts with OG1 on Thr 73 and O on Ala 69 of HLA-A2. WATGEN predicted
WAT12 is close to the X-ray water, at only 0.42 Å away from X-ray HOH 48.

18

2.3.2. WATGEN predicted water networks before and after peptide binding to MHC
This project intends to predict water networks of peptide-protein interface. There is no
guarantee of having a complete water network in an X-ray structure, and in many cases, no water
molecule is shown in a PDB file. To estimate the effect of peptide binding to MHC Class I, we
compared the water networks before and after peptide binding. To make such a comparison, the
water networks before and after peptide binding to MHC are placed in the same file. Water
molecules are paired together based on the distance before and after the binding. With the change
in position of water molecules before and after binding, the scores of the same water molecule
can also be different. The difference in water score is then presented by the colors assigned to the
water molecules in a PyMOL file.

Figure 2.3.2A WATGEN-predicted water network of a peptide-MHC class I complex.

In Figure 2.3.2A, only water molecules from the peptide-MHC class I complex run are
present. The colors of water molecules are determined by the energy of the water molecule. Red
molecules indicate that the energy of the water molecule in the complex is worse than the paired
water molecule in the protein only run. Green water molecules indicate that water molecules in
19

the complex are better energetically than the paired water molecules in the protein only run.
Yellow water molecules indicate similar energies to the MHC class I only run. White water
molecules mean that a paired water molecule could not be found in the protein only run. Since in
this run, the peptide is included, displaced water is not shown when the peptide binds to the
MHC cleft. To acquire the displaced water, the peptide is placed back in the MHC class I only
run. We can then record the water molecules that overlap with the peptide or are in close
proximity to the peptide, and then consider these water molecules as “displaced” by the binding
of peptide to the MHC cleft. These water molecules are shown in pink in Figure 2.3.2B.

Figure 2.3.2B WATGEN predicted network of MHC class I only.

In Figure 2.3.2B, water molecules from the MHC class I run are presented. The green
waters are better energetically than the complex, and red are worse energetically than in the
complex. Pink water molecules were replaced once the peptide binds to the MHC cleft. The
green and red colors in Figure 2.3.2B are opposite to Figure 2.3.2A, because colors are assigned
to indicate the differences in energy value.
20

Other than the number of displaced water acquired from the runs, the binding energy
difference can also be calculated by adding the individual scores of the water molecules in the
runs. A typical brief summary of run data is shown in the table below, as presented in a
“water.out” file.

Table 2.3.2A Overall water statistics on PDB:1S9W WATGEN runs.
Struct nwat score swb dwb swh dwh hbp int non
PL_1s9w_waters.pdb 344 -2895 123 124 0 0 0 0 97
PO_1s9w_waters.pdb 387 -5500 152 156 0 0 0 0 79

In Table 2.3.2A, two different WATGEN runs are shown: “PL” protein ligand, is the first
WATGEN run on the peptide-MHC class 1 complex. “PO” – protein only, is the second run with
MHC class I only. “nwat” is the number of water predicted by WATGEN in the interface. The
score is an arbitrary number which helps to quantify the difference between before and after
peptide binding to MHC class I. The lower the value, the more favorable the water network.
“swb” is an abbreviation of “single water bridge”, and “dwb” is an abbreviation of “double water
bridge”. They both contribute to the scoring of the water networks. This table only shows a
summary of the predicted energy; there are many more types of interactions considered in
WATGEN.

21

Table 2.3.2B Some other water statistics involved in the WATGEN score calculation.
(PDB:1S9W WATGEN runs).

22

2.3.3 Displaced water and its potential in docking
Figure 2.3.3.1 Number of displaced water molecules vs. Predicted IC50 .

Figure 2.3.3.2 WATGEN displaced water vs. EXSAN scores.

0
5
10
15
20
25
30
35
40
45
50
0 1000 2000 3000 4000 5000 6000
Displaced water vs. Predicted IC50
y = -0.001x + 26.091
R² = 0.1043
0
5
10
15
20
25
30
35
40
45
50
-12000 -10000 -8000 -6000 -4000 -2000 0
Displaced water vs. EXSAN Scores
23

2.4 Discussion
WATGEN is a powerful tool for building water networks of peptide-MHC class I
interface with fair accuracy in a short amount of running time - less than 30 seconds on a
personal desktop for each run. It provides a great opportunity to estimate water networks for
interfaces that do not have water molecules available in the binding site. Having such water
networks can potentially improve docking results for interfaces that have extensive involvement
of water networks. Different peptide conformations can result in different water networks that
have different energy, and such information can help determine whether a certain conformation
of the peptide is more favorable for binding. Although WATGEN is not perfect in terms of the
scoring system, it shows promising data to work on and improvements can be made by fine-
tuning the parameters to work on different interfaces.

24

Chapter 3: Docking peptides in MHC class I molecules: EXSAN
3.1 Introduction
EXSAN (EXplicit Solvent ANchored docking) is a software developed by Dr. Ian
Haworth’s Lab in University of Southern California School of Pharmacy for predicting the
peptide structure in a given X-ray structure of target protein. It utilizes explicit water molecules
during the prediction process, which increases the accuracy of the peptide binding. Previously,
EXSAN has been applied to PDZ domains and peptide-MHC class II complex. In this study, the
focus is on peptide-MHC class I complexes. The X-ray structures were retrieved from the
Protein Data Bank. Protein chains and peptide chains are identified and saved separately for the
runs.

Figure 3.1 An overview of EXSAN. (Brill, 2018)

25

3.2 Methods
Simplified version of an EXSAN run: Generating water structures in the p-MHC
interface: 1. Anchoring of the terminus 2. Generating poses of aa in the peptide (backbone to side
chain) 3. Culling the less favorable poses and repeat step 3,4 until reaching to the end 4.
Evaluating and selecting the best poses as the most favorable of the p-MHC pair.

Figure 3.2 Flow Chart of EXSAN. The phases of the program are color-coded as
initiation (red), growth (orange), solvation (green), scoring (blue), culling (purple), and
consensus (black). (Brill, 2018)

This flow chart shows the step for EXSAN to run. Different processes are colored:
initiation (red), growth (orange), solvation (green), scoring (blue), culling (purple), and
consensus (black). In the new version of EXSAN, the calculation of water network is only done
at the beginning of the program. This drastically increases the running speed with minimal loss
in the pose quality. The scores of the poses are used to evaluate how poses perform in the
complex. Taking multiple factors on this process, EXSAN is capable of narrowing the poses for
the next step. The culling step can be determined upon users’ preference. For amino acids that
26

don’t have particularly strong interactions, the user can keep more poses in that step so that
EXSAN can have more poses to work within the later stage. Upon the last addition of an amino
acid, EXSAN groups the similar poses as consensus poses.
The initial process includes anchoring the position of the first amino acid as a starting
point. This can be the N-terminus or C-terminus of the peptide, or an internal amino acid. To
anchor the atom, a measurement of the distance and torsional angle of that atom to the protein
are required and entered in EXSAN. It is recommended to use atoms that have strong interaction
to the protein for optimal performance. Once the anchored atom is set, EXSAN then extends the
peptide by adding backbone atoms first, followed by the side chain. Only one residue is added at
a time. EXSAN generates different conformations of the newly added residues by using various
rotatable bonds. The default setting for obtaining the rotatable bond was used in this work.
In the growth phase, an external program, TMD, is used to generate poses that are
permutated in different torsional angles with default increment of 30˚, which can be changed
according to the users’ preferences. Newly generated poses are examined to determine whether
they are too close to the target protein or in a conformation that conflicts with atoms on the
peptide. Such poses will be eliminated before going forward to the next extension phase.
EXSAN also eliminates poses that have less favorable carbonyl oxygen geometry and keeping
the optimum poses with carbonyl oxygen and backbone amide hydrogen interacting with
hydrogen bond donors and hydrogen bond acceptors, respectively.
EXSAN integrates the explicit water in the complex by calculating the score of
displacement of water by the peptide binding to the protein. This scoring method mimics the
actual binding of peptide to the MHC class I pocket. Depends on which water got displaced by
27

the peptide, the score is different. The summation of all the scores is assigned to that particular
pose for further analysis.
Scoring parameters are an important part of EXSAN: Scoring parameters in EXSAN
program. These parameters determine the score of the poses, and EXSAN is able to cull the bad
poses based on the score. There are three major categories to consider in EXSAN binding score:
1. Interactions between peptide and protein, 2. Water mediate interactions, and 3. Other
unfavorable position. Under each category, there are several parameters to considered.
There are four subcategories under the interactions between peptide and protein. Clash
limit is the limit that EXSAN considers too close for two heavy atoms between protein and
peptide to co-exist within this range. The poses that has the two heavy atoms within this distance
is considered very unfavorable and is eliminated. Salt bridges: it is considered salt bridge when
two atoms in the protein and peptide are from 1.7-4.4 Å (direct salt bridge) or 4.4-8.0 Å (water-
mediated salt bridge). Direct Hydrogen Bond: it is a score assigned to backbone-backbone
hydrogen bond when the distance between hydrogen bond donor and hydrogen bond acceptor are
within reasonable range and meet the appropriate angle. Hydrophobic contact: it is a score
assigned to peptide carbon/sulfur atoms of hydrophobic residue within 2.3-4.5 Å of another
protein carbon/sulfur atoms of hydrophobic residue.
Water mediated interactions also have four sub-categories. Water hydrophobic score is
the score assigned to water molecules that are within 3.5 Å of carbon/sulfur atoms of the
hydrophobic residues of either the peptide or proteins. Water ring score is the score assigned to
the water molecules within 3.5 Å of a aromatic carbon atom in the residue. Water salt bridge
score is a penalty score that caused by the displacement of water molecules that are previously
28

interacting with a charged residue. Water HB score is the score assigned to the displaced water
that are hydrogen bonding with the protein residues.
Other unfavorable positions are also considered in EXSAN as penalty in calculating the
EXSAN binding score. Bulk Facing Side Chain (SC) is the score that assigned to the side chain
with polar atoms facing toward the bulk solution instead of the center of the protein. Bad Pocket
score is the score assigned to hydrophilic side chain that point towards the hydrophobic pocket of
the protein. Electrostatic Repulsion is the score assigned to similar charges within 4.4 Å.
The scoring parameters play important role in curating the more desirable poses in the
runs. The more negative scored a pose get, the better the ranking of that pose. Since it is possible
to have poses that score extremely well by random chance, but, that are not possible. EXSAN
uses consensus poses to make sure that the well scored poses are attainable.
Depending on how accurate a user wants EXSAN to estimate, user can change the
accuracy in sacrifice of increasing running time. Users can also define the total amount of poses
kept by EXSAN after every additional residue is added and culled. Theoretically, the more poses
it keep, the more accurate the runs, because in the early stage, the peptide may not be very
energetically favorable. Keeping a small number of poses may affect the quality of the final
peptide RMSD. On the other hand, keeping too many poses after every stage may increase the
computational time by a significant amount. EXSAN is also capable of getting rid of poses that
have residues too far away from the binding pocket. This is made possible by restricting residues
within a certain distance of atoms of the protein.
To demonstrate the ability of EXSAN in predicting the peptide conformation on MHC
class I cleft. X-ray structures were retrieved from PDB. We presume that EXSAN score is
sophisticated enough to detect the affinity of peptide to MHC class I.
29

3.3 Results and discussion
3.3.1 Validation of EXSAN
Figure 3.3.1A EXSAN binding scores and experimental IC50.

Figure 3.3.1B EXSAN binding scores and experimental IC50.

y = 1.2498x + 8811.5
R² = 0.2666
-4000
-2000
0
2000
4000
6000
8000
10000
12000
14000
16000
-9000 -8000 -7000 -6000 -5000 -4000 -3000
IC50 vs EXSAN Binding scores
y = -0.0011x + 0.3891
R² = 0.8164
0
2
4
6
8
10
12
-9000 -8000 -7000 -6000 -5000 -4000 -3000
pIC50 vs EXSAN Binding scores
30

Figure 3.3.1A&B show the relationship between the experimental pIC50 for the complexes in
Table 2 and the EXSAN binding score. pIC50 is the logarithmic form of IC50. The larger the
pIC50, the higher the binding affinity of the peptide to the MHC class I.
The EXSAN binding scores are based on application of the parameter in Table 1 in the
supplement to the X-ray structures in Table 2. The results show a strong relationship between the
experimental and calculated values. This allows an IC50 to be predicted for a structure built in
EXSAN.

Figure 3.3.1C Interface number ranked by the EXSAN binding scores.

Figure 3.3.1C is used to rank the different binding scores yield by EXSAN while X-ray
structures were used in generating the EXSAN scores. The goal is to see whether EXSAN is
capable in generating logical score to be used in docking. These 50 different interfaces are rank
-8500
-7500
-6500
-5500
-4500
-3500
-2500
-1500
-500
1 6 11 16 21 26 31 36 41 46
EXSAN Binding Score
Interface Number
31

by the EXSAN binding score because it is easier to observe which parameter play significant
roles in determining the EXSAN binding score.
Figure 3.3.1D Number of Salt Bridges vs. Interface Rank by EXSAN scores.

On Figure 3.3.1D, the number of salt bridges between peptides and interface number are
presented.

0
2
4
6
8
10
1 6 11 16 21 26 31 36 41 46
Number of Salt Bridges
Interface Number
32

Figure 3.3.1E Number of Hydrophobic interactions vs. Interface Rank by EXSAN scores.

0
5
10
15
20
25
30
1 6 11 16 21 26 31 36 41 46
Number of
Hydrophobic Interactions
Interface Number
33

Figure 3.3.1F Number of hydrophobic interactions recognized in EXSAN and the interface
number (Red). Number of displaced water and the interface number (Blue).

On Figure 3.3.1F, the top of this figure shows the number of displaced water by the
peptide and the bottom of this figure shows the EXSAN energy of displaced water, which is a
parameter depends on the number of water molecules displaced and the location of the water
molecules. In general, the more water displaced by the peptide, the WATGEN score tend to be
lower, which cause EXSAN binding score to be even lower – indication of better binding.

-80
-60
-40
-20
0
20
40
60
80
-3000
-2000
-1000
0
1000
2000
3000
1 6 11 16 21 26 31 36 41 46
Number of Displaced Water
EXSAN Energy of Displace Water
Interface Number
34

Figure 3.3.1G EXSAN hydrogen bond energy and interface number (Red). Number of hydrogen
bonds and the interface number (Green).

On Figure 3.3.1G, the hydrogen bond energy (bottom red) and number of hydrogen
bonds (top green) are presented according to the ranked interface number. We presume that the
more hydrogen bonding between peptide and MHC class I or water molecules in the cleft, the
better the affinity. This figure is organized this way because scores are assigned to the hydrogen
bond present in the binding site, and it is calculated as a parameter of the EXSAN binding score.
The EXSAN hydrogen bond energy is a measure of how many hydrogen bonds are there
between the peptide and protein, and this information is pass to EXSAN to calculate the binding
score. Note: the y-axis of hydrogen bond energy goes from 0 to -4200 and the y-axis of number
of hydrogen bonds goes from 0 to 12.
-13
-8
-3
2
7
12
-4200
-3200
-2200
-1200
-200
800
1800
2800
3800
1 6 11 16 21 26 31 36 41 46
Number of Hydrogen Bonds
EXSAN Hydrogen Bond Energy
Interface Number
35

Figure 3.3.1H EXSAN Additional Energy vs. ranked interface number.

On Figure 3.3.1H shows a table on all the other measurements of EXSAN score.

-4500
-4000
-3500
-3000
-2500
-2000
-1500
-1000
-500
0
1 6 11 16 21 26 31 36 41 46
EXSAN Additional Energy
Interface Number
36

3.3.2. X-ray structure vs. EXSAN predicted peptide conformation
The primary way to examine EXSAN is to compare the predicted structure to the X-ray
structure. We have ran several interfaces to compare the predicted peptide conformation to the
X-ray crystallography of peptide-MHC class I complex.

Case 1: Docking Influenza A matrix protein to HLA-A2. PDB: 1HHI (Madden, 1993)
1HH1 is retrieved from PDB file. It’s a complex of influenza A virus matrix protein to
the human class I MHC molecule HLA-A2. The peptide sequence is GILGFVFTL. In this run,
we anchored the N-terminus of the peptide as a starting point. In the experiment set up, we also
limit the position of C-alpha of the backbone near to the original X-ray structure about +/- 1 Å.
In the experiment set up, only 10,000 poses are kept in each side chain addition step to reduce
the calculation time.

37

Table 3.3.2A Poses generated by TMD in the growth phase in EXSAN.
Sequence Action
Poses
Count Remark
G TMD 1
GG TMD 53
GI TMD 162
GI waterCull 49 Top 10000
GIG TMD 689
GIL TMD 1387
GIL waterCull 423 Top 10000
GILG TMD 3917
GILG waterCull 1166 Top 10000
GILGG TMD 8150
GILGF TMD 19457
GILGF waterCull 2863 Top 10000
GILGFG TMD 32639
GILGFV TMD 76088
GILGFV waterCull 9336 Top 10000
GILGFVG TMD 57969
GILGFVF TMD 95284
GILGFVF waterCull 10000 Top 10000
GILGFVFG TMD 46231
GILGFVFT TMD 105640
GILGFVFT waterCull 10000 Top 10000
GILGFVFTG TMD 13329
GILGFVFTL TMD 112146
GILGFVFTL waterCull 9996 Top 10000

On Table 3.3.2., it shows the poses of Growth Phase and the Culling Steps. In the early
steps, there aren’t that many poses to work on. As the peptide grow towards the C-terminus,
there are more poses kept by EXSAN and take longer time to score all the poses and cull the
less-favorable poses. TMD is an external program used by EXSAN to generate poses and add
atoms on the previous step. In Table 3.3.2. Only TMD steps are shown to simplify the table.
There are multiple culling steps following the TMD steps to reduce the unfavorable poses,
38

including Atom Cull, Water Cull, Extend Cull, Backbone Cull and Distant Cull. In the culling
phase, to further reduce the running time, only the top 10,000 favorable poses were kept for the
later stage.
From the first step. The peptide was grown from the anchored glycine and then attached
another glycine to C-terminus of the first glycine. After the first step, isoleucine was then added
to replace the side chain of glycine. After the final step, the Tier 1 result is shown in Fig. x. It is
similar to the X-ray structure with RMSD backbone score of 2.322 Å and RMSD side chain
score of 8.338 Å. The overlay between the X-ray structure (Green) and the best predicted
result(Purple). EXSAN is capable in predicting the position of Phe5 and Phe7 that’s similar to
the X-ray structure. The prediction of Leu3 does not perform well. Predicted Gly4 is similar to
X-ray Gly4 position as well. Closer to the C-terminus, predicted Thr8 and Leu9 are similar to the
X-ray structure as well. The prediction of Leu9 could have been better, but the prediction is not
accurate enough to figure out the exact conformation.

39

Figure 3.3.2A X-ray crystallography (purple) of GILGFVFTL and the predicted peptide
conformation by EXSAN (green). The amino residue name and number are labeled in orange
from N-terminus to C-terminus.

Figure 3.3.2B Predicted peptide conformation by EXSAN and the displaced water molecules

40

Fig. 3.3.2B shows that the water molecules before binding got displaced by the binding
of the peptide. Depending on the types of water molecules, different displaced water molecules
scored differently. For example, it is a favorable conformation to displace water molecules in the
hydrophobic pocket for gaining extra enthalpy energy to the binding. On the other hand, it is less
favorable to strip away a hydrogen bonding water and replaced that water molecule with a
hydrophobic residue after binding. The water displacement score helps EXSAN to predict the
better conformation of the docking.

41

Figure 3.3.2C X-ray crystallography of GILGFVFTL and the best predicted result from EXSAN
with MHC molecules in the background.

From Figure 3.3.2C, the peptide sits reasonable within the binding pocket of HLA-A2
molecule. There isn’t any part that is perturbing outside of the pocket.

42

Figure 3.3.2D Interaction between the peptide and Arg 97, Tyr 116 on HLA-A2.

43

Case 2: Docking Influenza A matrix protein to HLA-A2. PDB: 2VLR (Ishizuka et al., 2008)
Compare EXSAN runs with and without the presence of TCR. From this group of
EXSAN runs, we observe differences on the accuracy of predicting peptide conformation by
EXSAN. Since TCR on the top of the peptide can provide extra constraint on peptide, it is
expected to see higher accuracy on the peptide structure. In the prediction result, the run without
TCR has higher accuracy (lower RMSD) than the run with TCR present.

Figure 3.3.2E X-ray structure of Influenza A matrix protein to HLA-A2 With TCR present.
Left: structure of TCR-peptide-MHC complex. Right: zoom in structure of the TCR-peptide-
MHC complex with TCR on the top, peptide in the middle, and MHC Class I on the bottom.

44

Figure 3.3.2F EXSAN prediction of peptide conformation. X-ray structure (grey) vs. TCR run
(green) vs. no TCR run (orange). Amino acid residues and their positions are in yellow. TCR is
on the top and MHC class I is at the bottom.
Two EXSAN runs on 2VLR with exact settings were examined, and the result of peptide
conformation is compared to the X-ray structure with backbone RMSD score and overall RMSD
score. The run with TCR present shows lower backbone RMSD () than the run without TCR
present. Structurally, Phe 5 on green resembles similar position to the grey conformation, while
the backbone of orange bend toward the other side of the grey structure. Green also share similar
conformation on Phe 7. Having TCR present in the run will help EXSAN with higher accuracy
since the binding groove is more confined.
45

Figure 3.3.2G EXSAN prediction of the peptide conformation with the presence of the TCR.

Figure 3.3.2H EXSAN prediction of the peptide conformation without the presence of the TCR.
46

Figure 3.3.2I EXSAN prediction of the peptide conformation without the presence of TCR and
without running WATGEN on every amino acid residue addition process (Purple – Backbone
RMSD: 1.53). X-ray structure (grey).

47

Figure 3.3.2J EXSAN prediction of the peptide conformation without the presence of a TCR
and without running WATGEN on every added amino acid residue (Purple – Backbone RMSD:
1.53). WATGEN running on every step (Green – Backbone RMSD: 3.07). X-ray structure
(grey).

48

Table 3.3.2B Comparison of different EXSAN setting and the RMSD scores.
PDB-
Code Allele Peptide
TCR in
EXSAN WATGEN
RMS -
backbone
RMS -
overall
2VLR
HLA-
A2 GILGFVFTL Y Every step 3.07 7.29
2VLR
HLA-
A2 GILGFVFTL N Every step 5.54 17.64
2VLR
HLA-
A2 GILGFVFTL Y Start step 1.53 4.31

From Table 3.3.2B¸ it shows three different EXSAN runs on the same interface using
different setups. Depending on whether WATGEN is executed in every additional steps, the
predicted structure have different accuracy.
Table 3.3.2C Brief summary of other runs conducted.
PDB Allele Sequnce (aa)
Epitope
Source
TCR WATGEN
RMSD-
BB
RMSD
2BST
HLA-
B2705
SRYWAIRTR
Influenza A
Nucleoprotein
N Start step 1.58 19.07
3CDG
HLA-
E
VMAPRTLFL
Human
CD94/NKG2A
Y Start step 1.31 4.23
3CDG
HLA-
E
VMAPRTLFL
Human
CD94/NKG2A
N Start step 6.84 21.11
2X4O
HLA-
A2
KLTPLCVTL
HIV-1
envelope
peptide
N Start step 0.97 4.27
3BUY
HLA-
H2-D
LSLRNPILV
Epitope of
PB1-F2
N Start step 1.52 4.93

Table 3.3.2C shows some other peptide-MHC class I docking results from EXSAN. In
the case of 3CDG run, the inclusion of T-cell receptor in the run shows a big increase in both
RMSD-backbone and overall RMSD compare to the run without the T-cell receptor. In this case,
the inclusion of the T-cell receptor restrains the space available for the peptide to move in the
MHC cleft, which can eliminate the possible poses in EXSAN significantly. Once the peptide
49

poses meet the “clash limit” of 1.7 Å to either the MHC class I protein or the T-cell receptor,
those poses will be eliminated by EXSAN. With the extra restrain by the T-cell receptor,
EXSAN is capable of finding more interactions between the peptide and the T-cell receptor,
which also help predicting the favorable peptide poses compare to the runs with MHC class I
only.

50

Chapter 4: Evaluation and Conclusion
4.1 Further study and Improvements
EXSAN is highly customizable, but it isn’t easy to use for the docking. It will be a great
idea to build a graphic user interface for easier setup for the docking process. Currently, users
need to prepare several files and change the setting manually before the docking, for example
user need to manually input the peptide and target protein chains before docking. EXSAN also
lacks the ability to run in a batch of runs. Building a user friendly interface can help EXSAN to
perform multiple runs, especially screen a large quantities of peptide-proteins combination
feasible.
Although EXSAN can predict peptide structures on shorter peptide in an efficient and
accurate way, the running time can increase significantly on longer peptides which have more
than ten amino acids. This can possibly be reduced by limiting the poses kept after each addition
of amino acid to peptides, but it will decrease the overall accuracy of docking results since some
energetically favorable poses might get eliminated at the early stage. Another way to improve the
docking result is to build the peptide separately from C-terminus and N-terminus separately, and
then join these two parts together in the middle. This setup could increase the accuracy because
in peptide-MHC class I complex, both C and N-termini of the peptide have relatively strong
interactions to the MHC class I protein.
In this project, there is no mutation on MHC class I proteins or peptides, which can be
further studied to include in silico simulation on mutating either proteins or peptides to examine
the difference in binding caused by the mutation. This approach can screen on a large number of
possible mutations in either peptide or MHC class I. It’s costly and time-consuming to obtain
certain mutated MHC class I, but with this in silico approach, such screening can be done within
hours and with low expenses.
51

4.2 Conclusion
WATGEN is a powerful tool in modeling the water networks of peptide-MHC class I
complexes efficiently and effectively, especially in estimating water molecules that have key
interactions with peptide or target proteins. WATGEN can process large quantities of requests in
an efficient and effective way. Having such an algorithm in building water networks can help
improve peptide-protein docking, especially for those interfaces that have complicated water
networks involved in between peptide and protein. It can help docking software to optimize the
peptide conformation by considering the effect of water in the interface. A minor change in
peptide conformation can cause a major change in energy calculation and eventually contribute
to the optimization of the peptide conformation.
Integrating WATGEN in the docking process, EXSAN is capable of predicting peptide-
MHC class I complex in reasonable running time and accuracy. The inclusion of water networks
in determining peptide conformation plays an important role in accurately choose the best
conformation from thousands to millions of potential conformations. Although EXSAN has
room to improve on user interface and running time, it is showing promising docking results with
high similarity to the X-ray structures.
Different from traditional in vivo and in vitro approach in biological/pharmaceutical
studies, in silico approach have the benefits of less cost and higher efficiency. Although in silico
approach is unlikely to replace in vivo or in vitro studies any time soon, WATGEN and EXSAN
are showing great potential in early drug discovery and screening experimental subjects from
large database.

52

Supplemental Data
Table 1. EXSAN Binding Scores and Experimental IC50 Values.
Interface Gene allele Ligand sequence
EXSAN Binding
Energy IC 50/nm pIC 50
3QFD HLA-A2 AAGIGILTV -4471.69913 6955 5.157703
2GT9 HLA-A*02:01 EAAGIGILTV -4631.551414 14560 4.836839
5U98 HLA-B*57:01 VTTDIQVKV -4867.08529 2000 5.69897
3REW HLA-A2 CLGGLLTMV -4890.172066 2500 5.60206
4NNY HLA-A-02 RQASLSISV -4950.000937 284.5 6.545918
3UPR HLA-B*57:01 HSITYLLPV -5066.543727 700 6.154902
1JF1 HLA-A*02:01 ELAGIGILTV/AAGIGILTVI -5166.268183 5555 5.255316
2X4S HLA-A*02:01 AMDSNTLEL -5413.727499 122 6.91364
3GIV HLA-A2 SLFNTVATLY -5419.231991 131 6.882729
3I6L HLA-A*2402 QFKDNVILL -5511.739671 1130 5.946922
3LKP HLA B*3501 LPFDKSTIM -5590.951502 1500 5.823909
5B39 HLA-B*57:01 LSSPVTKSF -5617.679466 389 6.41005
3UTQ HLA-A*0201 ALWGPDPAAA -5668.087625 1008 5.996539
3LKS HLA B*3501 LPFEKSTVM -5758.864421 1100 5.958607
1BD2 HLA-A*02:01 LLFGYPVYV -5817.069326 800 6.09691
1T20 HLA-A2 SLYNTIATL -5891.951114 620 6.207608
5EU3 HLA-A2 YLEPGPVTA -5900.806603 95 7.022276
3FT3 HLA-A2 VLHDDLLEA -5901.792251 30 7.522879
3GSO HLA-A2 NLVPMVATV -5969.515124 45 7.346787
1HHI HLA-A*02:01 GILGFVFTL -6104.83173 1200 5.920819
2X4O HLA-A*02:01 KLTPLCVTL -6108.222335 102 6.9914
3MRM HLA-A2 KLVALGINAV -6197.196442 36 7.443697
1HHJ HLA-A*02:01 ILKEPVHGV -6238.4628 192.3 6.716021
3RL1 HLA-A*0301 AIFQSSMTK -6284.400823 10 8
1A1M HLA-B*53:01 TPYDINQML -6288.24982 30.38 7.517412
5EU6 HLA-A YLEPGPVTV -6458.107782 45.5 7.341989
1QEW HLA-A*02:01 FLWGPRALV -6466.063553 31.3 7.504456
3BO8 HLA-A1 EADPTGHSY -6481.85437 44 7.356547
2FO4 H2-Kb SAPDFRPL -6486.232077 60 7.221849
3GSN HLA-A2 NLVPMVATV -6531.165396 45 7.346787
2HN7 HLA-A-1101 AIMPARFYPK -6596.399598 31 7.508638
4HWZ HLA-A68 AIFQSSMTK -6604.307731 144 6.841638
3V5H HLA-A2.1 KVAEIVHFL -6710.317529 66 7.180456
5TEZ HLA-A2 GILGFVFTL -6805.975808 3.5 8.455932
5T6Z HLA-B*57:01 TSTLQEQIGW -6830.395137 6.96 8.157391
53

3MRG HLA-A2 CINGVCWTV -6896.135151 55 7.259637
5HGH HLA-A*24:02 RYPLTFGWCF -7004.221898 17 7.769551
3DX6 HAL-B*44:02 EENLLDFVRF -7081.155123 1.58 8.801343
5BRZ HLA-A1 EVDPIGHLY -7197.530038 4.6 8.337242
1FZM H2-K RGYVYQGL -7270.143176 1 9
3DX7 HLA-B*4403 EENLLDFVRF -7312.615795 11.2 7.950782
3OXR HLA-A*02:06 FLPSDFFPSV -7430.728695 3.1 8.508638
5EUO HLA-A2 GILGFVFTL -7539.937417 3.5 8.455932
5W1V HLA-E VMAPRTLIL -7661.762661 0.5 9.30103
3RWC
Mamu-B*17-
IW9 IRYPKTFGW -7709.028561 5.57 8.254145
3V5K HLA-A2.1 KVAELVWFL -7718.491703 7 8.154902
2BNQ HLA-A SLLMWITQV -8009.953265 1.2 8.920819
4G9D HLA-B*27:05 KRWIILGLNK -8111.838299 0.22 9.657577
2J8U HLA-A*02:01 ALWGFFPVL -8189.474817 2.7 8.568636
1MI5 HLAB8 FLRGRAYGL -8332.391301 1 9

54

References
Antes, I., Siu, S.W.I., and Lengauer, T. (2006). DynaPred: A structure and sequence based
method for the prediction of MHC class I binding peptide sequences and conformations.
Bioinformatics 22, e16-e24.
Antunes, D.A., Devaurs, D., Moll, M., Lizée, G., and Kavraki, L.E. (2018). General Prediction
of Peptide-MHC Binding Modes Using Incremental Docking: A Proof of Concept. Scientific
Reports 8.
Bui, H.-H., Schiewe, A.J., and Haworth, I.S. (2007). WATGEN: An algorithm for modeling
water networks at protein-protein interfaces. 28, 2241-2251.
Ishizuka, J., Stewart-Jones, G.B.E., Van Der Merwe, A., Bell, J.I., McMichael, A.J., and Jones,
E.Y. (2008). The Structural Dynamics and Energetics of an Immunodominant T Cell Receptor
Are Programmed by Its Vβ Domain. 28, 171-182.
Kyeong, H.-H., Choi, Y., and Kim, H.-S. (2018). GradDock: rapid simulation and tailored
ranking functions for peptide-MHC Class I docking. Bioinformatics 34, 469-476.
Madden, D. (1993). The antigenic identity of peptide-MHC complexes: A comparison of the
conformations of five viral peptides presented by HLA-A2. Cell 75, 693-708.
Webb, A.I., Dunstone, M.A., Chen, W., Aguilar, M.I., Chen, Q., Jackson, H., Chang, L., Kjer-
Nielsen, L., Beddoe, T., McCluskey, J., et al. (2004). Functional and Structural Characteristics of
NY-ESO-1-related HLA A2-restricted Epitopes and the Design of a Novel Immunogenic
Analogue. 279, 23438-23446.
Brill, D. (2018). EXSAN: Explicit Solvent Anchored Fragment-Base Docking. University of
Southern California.

Abstract (if available)

Abstract This thesis demonstrates the potential of computational methods to predict the peptide structure in peptide-MHC class I complexes. There are two parts of the thesis: 1. modeling the water network in the peptide-MHC class I complexes using WATGEN

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Solvation as a driving force for peptide docking to the major histocompatibility complex (MHC) class II molecules

PDF

Structure-based computational analysis and prediction of TCR CDR3 loops in the TCR-peptide-MHC complex using solvation parameters and peptide molecular dynamics.

PDF

Prediction of peptides in formation of MHC class I - peptide - TCR complexes using molecular models and artificial intelligence

PDF

Computer modeling of protein-peptide interface solvation

PDF

Structural prediction of MHC-peptide-TCR interactions: potential for vaccine design

PDF

Modeling of water molecules in protein-ligand binding

PDF

Computational model for predicting ionic solubility

PDF

Computational models for drug design and delivery

PDF

EXSAN: explicit solvent anchored fragment-base docking

PDF

Molecular modeling of cyclodextrin interactions with proteins

PDF

Computational analysis of drug complexes with beta-cyclodextrin

PDF

Y-shaped DNA based pMHC nonamers for detecting low-affinity T cells

PDF

Towards DNA-directed assembly of pMHC multimers for detection of low-affinity T cells

PDF

Characterization of the interaction of nucleotide-binding oligomerization domain, leucine-rich repeat and pyrin domain-containing protein 12 (Nlrp12) with hematopoietic cell kinase (Hck)

PDF

Regulation of miR21, miR221, CXCL5 and OPN in liver cancer

PDF

Inhibition of monoamine oxidase A and histone deacetylase inhibitors: computational prediction of ligand binding

PDF

Site-specific PEGylation of recombinant immunotoxin αFAP-PE38 for half-life extension

PDF

Assessment of the immunomodulatory potential of G. lucidum using computational and in vitro methods

PDF

Computational prediction and analysis of protein-RNA interfaces

PDF

Computational prediction of peptide-MHC class I binding interactions

Asset Metadata

Creator Lai, Jiawei (author)

Core Title Computational modeling of solvation and docking of peptide-MHC class I

School School of Pharmacy

Degree Master of Science

Degree Program Pharmaceutical Sciences

Publication Date 08/16/2019

Defense Date 07/21/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag computational modeling,docking,Immune System,MHC,OAI-PMH Harvest,peptide-protein complex,solvation

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Haworth, Ian (committee chair), Romero, Rebecca (committee member), Xie, Jianming (committee member)

Creator Email jiaweila@usc.edu,laijw2048@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-211462

Unique identifier UC11663124

Identifier etd-LaiJiawei-7805.pdf (filename),usctheses-c89-211462 (legacy record id)

Legacy Identifier etd-LaiJiawei-7805.pdf

Dmrecord 211462

Document Type Thesis

Format application/pdf (imt)

Rights Lai, Jiawei

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA