University of Southern California Dissertations and Theses

Inhibition of MAO-A by Dual MAO-A/HDAC inhibitors: in silico approach for ligand binding and affinity prediction

(USC Thesis Other)

Inhibition of MAO-A by Dual MAO-A/HDAC inhibitors: in silico approach for ligand binding and affinity prediction

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
Inhibition of MAO-A by Dual MAO-A/HDAC Inhibitors:
In silico Approach for Ligand Binding and Affinity Prediction

by

Sawanee Joshi

Thesis Presentation to the
FACULTY OF THE USC ALFRED E. MANN SCHOOL OF PHARMACY AND
PHARMACEUTICAL SCIENCES
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(PHARMACEUTICAL SCIENCES)
August 2023
ii
Table of Contents
List of Tables ----------------------------------------------------------------------------------------------iii
List of Figures----------------------------------------------------------------------------------------------iv
Abstract-----------------------------------------------------------------------------------------------------vi
Chapter 1----------------------------------------------------------------------------------------------------1
Introduction-------------------------------------------------------------------------------------------------1
Monoamine Oxidase---------------------------------------------------------------------------------------1
Histone Deacetylase(HDAC)------------------------------------------------------------------------------2
Dual Inhibitors:MAO-A/HDAC--------------------------------------------------------------------------2
Chapter 2-----------------------------------------------------------------------------------------------------4
TMD Runs and Solvation Free Energy Runs-----------------------------------------------------------4
TMD Setup: Generating Clorgyline-Based Inhibitor .zmat files-------------------------------------5
TMD Molecular Docking----------------------------------------------------------------------------------8

Chapter 3-----------------------------------------------------------------------------------------------------17
Machine Learning Model----------------------------------------------------------------------------------17

Chapter 4-----------------------------------------------------------------------------------------------------42
Application of Model---------------------------------------------------------------------------------------42
References----------------------------------------------------------------------------------------------------45
iii
List of Tables
Table 1: Torsion angles tested using TMD exploring different possible conformations of 440
and 441 Model 1-5 , set 1, set2--------------------------------------------------------------------------10
Table 2: Experimental MAO-A IC50 data using GL26 cells----------------------------------------16
Table 3: Nodes used and their functions----------------------------------------------------------------19
Table 4: Functions of categories used for the model--------------------------------------------------20
Table 5: Combinations of 19 different models---------------------------------------------------------21
iv
List of Figures
Figure 1: General drug design structure of MAO A-HDAC inhibitors-------------------------------3
Figure 2: Structure of the Dual MAO-A/HDAC inhibitors that are analyzed-----------------------4
Figure 3: Raw .zmat file of 440 using torsions and bond angles of clorgyline in the crystal
structure-------------------------------------------------------------------------------------------------------6
Figure 4: Raw .zmat file of 441 using torsions and bond angles of clorgyline in the crystal
structure-------------------------------------------------------------------------------------------------------7
Figure 5: Structures of protein models 1-5 with strands created using MODELLER--------------9
Figure 6: General KNIME workbench-------------------------------------------------------------------17
Figure 7: Complete KNIME (workflow) model used--------------------------------------------------18
Figure 8: Only HB and HF---------------------------------------------------------------------------------22
Figure 9: Addition of RB bonds –-------------------------------------------------------------------------23
Figure 10: Addition of MW--------------------------------------------------------------------------------24

iv
Figure 11: Addition of ASWB-----------------------------------------------------------------------------25
Figure 12: Addition of CSWB-----------------------------------------------------------------------------26
Figure 13: Addition of ADW-------------------------------------------------------------------------------27
Figure 14: Addition of TNLW-----------------------------------------------------------------------------28
Figure 15: Removal of MW--------------------------------------------------------------------------------29
Figure 16: Re-addition of MW-----------------------------------------------------------------------------30
Figure 17: Removal of both MW and LogP--------------------------------------------------------------31
Figure 18: Removal of ADW------------------------------------------------------------------------------32
Figure 19: Removal of Log P and ADW-----------------------------------------------------------------33
Figure 20: Removal of all solvation categories except LogP, HB,HF--------------------------------34
v
Figure 21: Removal of solvation parameters(ASWB, CSWB, ADW, TNMLW)------------------35
Figure 22: Readdition of solvation parameters, along with MW, logP and TMPW----------------36
Figure 23: The addition of another solvation parameter TMLW--------------------------------------37
Figure 24: Addition of TIMW-----------------------------------------------------------------------------38
Figure 25: Removal of TMPW----------------------------------------------------------------------------39
Figure 26: Removal of LogP and readdition of TMPW-----------------------------------------------40
Figure 27: Shows the percentage prediction or accuracy for each of the models------------------41
Figure 28: Model 15 used for MOA-A/HDAC dual inhibitor affinity prediction------------------42
Figure 29: Energetically most favourable position calculated for 440 was from set 2-------------42
Figure 30: Energetically most favourable position calculated for 441 was from set 2-------------43

vi
Abstract
Molecular docking is one of the most common and established methods in drug discovery. The
purpose is to predict and observe the interaction between small molecules and a protein at an
atomic level. This method is especially used for identification of novel molecules that may be
therapeutically important. However, the most molecular docking software is catered to
generating poses for reversible inhibitors. Generating poses or models for irreversible inhibitors
is less common.
The aim of this study is to model dual MAO-A/HDAC inhibitors that bind irreversibly to the
FAD cofactor in the MAO-A enzyme. In order to generate models, we used an in-house
software, TMD (Tethered Molecular Docking), that scanned potentially possible poses from the
library of inhibitors that was given as an input. To understand the energetic contributions of these
poses, solvation calculations were performed from another in-house software, Watgen5. It was
inferred from the poses of these complexes that the effects of solvation in the MOA-A binding
site are important in order to drive the ligand binding.
To obtain a more detailed evaluation of the binding energetics of MAO-A inhibitors, we
explored the behavior of the predicted and solvated poses in a machine learning model. This
model was constructed in KNIME based on data obtained previously for >9000 solvated protein-
ligand complexes. The model is still preliminary, but we show that it may be able to predict
affinity to within about 1 kcal/mol. This model was used to attempt a prediction of the affinities
of the tested ligands with MAO-A.
1
Chapter1: Introduction
Monoamine oxidase:
Monoamine oxidases (MAOs) are outer mitochondrial membrane proteins that catalyze the
oxidation of primary, secondary, and tertiary amines, including several neurotransmitters, to the
corresponding imines; the oxidized products are hydrolyzed non-enzymatically to the respective
aldehydes or ketones. However, this study is focused on MAO A and its correlaion to cancer.
The enzyme contains a covalently-bound flavin adenine dinucleotide (FAD) cofactor attached to
a cysteine via the 8α-methylene of the isoalloxazine ring, MAO A metabolizes serotonin (5-
hydroxytryptamine), norepinephrine, and dopamine. Inhibitors of the monoamine oxidases have
been used clinically for the treatment of depression, as well as Parkinson’s, Alzheimer’s and
other neurodegenerative diseases
1
. Recent research has shown the pivotal role of MAO A as a
novel cancer target; MAO A inhibitors inhibited the progression and growth of brain cancer
2,3

and prostate cancer
10
.These reports demonstrated for the first time that antidepressants can be
repurposed for cancer chemotherapy.
The active site of MAO A is surrounded mostly by hydrophobic and aromatic moieties and is
refered to as an aromatic cage. The FAD co-factor located in this cage is required for the
oxidative deamination of monoamine neurotransmitters. The binding cavity extends from the
cofactor to the loop containing residues 210-216. This flavone in the FAD is also a target of
MAO A inhibitors similar to the potent suicide substrate clorgyline, which selectively inhibits
MAO A by irreversibly binding to FAD via its alkyne and prevents further catalysis
5
. It has
been shown that MAO A induces epithelial-to-mesenchymal transition (EMT) through activation
of vascular endothelial growth factor (VEGF) and its co-receptor neuropilin-1. Therapies such as
these which target MAO A may represent a new opportunity for the treatment of patients with
recurrent prostate cancer.
5
Clorgyline is a FDA approved drug used as an antidepressant. Though
it is not proven as a possible anti cancer agent, an unexpected finding showed that MAO A is
highly expressed in EnZ-resistant prostate cancer cells and clorgyline can restore Enz sensitivity
to further suppress EnzR cell growth.
9
2
Histone Deacetylase (HDAC):
The process of histone acetylation and deacetylation are commonly studied epigenetic
modifications and play an interesting role in carcinogenesis and progression. Gene expression
regulation by modulating post-translational changes of histones has generated interest in the
development of various therapies that are capable of altering gene expressions.
5

HDACs are enzymes that catalyze the removal of acetyl functional groups from the lysine
residues of histones and non-histone proteins. There are 18 HDAC enzymes present in humans
that use zinc or NAD
+
dependent mechanisms to deacetylate lysine substrates. HDACs are
responsible for the removal of histone acetyl epigenetic modifications which regulate chromatin
restructuring and transcription. Deacetylation of non-histones controls diverse cellular processes.
HDAC inhibitors are now known to be potential anticancer agents and have shown promising
results for the treatment of many diseases.
7
Many HDACs have been reported to be
overexpressed in various malignancies, and therefore, various HDAC inhibitors have been
developed to treat these diseases. To state a few examples, HDAC class IIA is reported to play an
important role in gliomagenesis, progression, and invasion; HDAC1 and HDAC2 undergo
significant changes in gliomas; and HDAC6 is overexpressed in glioblastoma and has been
correlated with poor patient survival. Interestingly, it has been observed that these enzymes are
also overexpressed in glioma stem-like cells (GSCs) and HDAC inhibitors could reduce tumors
(in vivo) by increasing the cell cycle activity. However, the role of HDAC inhibitors in the
treatment of solid tumors, such as glioblastoma multiforme may still need to be strengthened.
5

Dual Inhibitors:MAO A-HDAC:
The current study aims to explore the different interactions that can occur between the set of
novel HDAC/MAO A dual inhibitors and MAO A (MAO isoenzyme) using an in silico ligand
binding approach. Here, the inhibitors have been docked to the MAO A active site to achieve
predicted structures of the inhibitors that have been covalently bound to the enzyme. This
process provides a general idea as to how these inhibitors would possibly bind.
These dual inhibitors contain a MAO-A targeting group and a HDAC targeting group, which are
connected together by a hydrophobic linker that varies in length and rigidity (Figure 1).
Previously conducted in vitro studies of these dual inhibitors have generated experimentally
3
determined affinities for MAO A.
5
These docked structures within the active site may aid in
designing of molecules with better delivery properties. Since these molecules are extremely
hydrophobic and MAO A contains a covalently bound FAD it can cause solvation issues. To try
and understand the solvation issue we need to predict affinity based on these generated docked
and solvated structures.To aid this problem we need a model, and this was achieved by
generating a machine learning model using KNIME. Using this computational approach, there is
room to slightly alter the structure and make it soluble while maintaining the excellent inhibitory
activity.

Figure 1: General drug design structure of MAO A-HDAC inhibitors.
The red circle shows the structure of hydroxamic acid, which is needed for HDAC inhibition. The green
circle shows a clorgyline-like compound needed for MAO A inhibition.
5
(Figure partially taken from S.
Mehndiratta paper).
4
Chapter 2: TMD Runs and Solvation Free Energy Runs
Need for Covalent Docking:
It is reasonable to assume that the dual inhibitors 440 and 441 (Figure 2) behave much like
clorgyline. Both clorgyline and dual inhibitors have the same required chemical structures that
are able to form a covalent bond with FAD and should be docked in a way that generates
irreversible complexes. Therefore, instead of using standard available docking software which is
more likely to favor non-covalent bonding, the solution to get useful results would be to use a
software that would be able to simulate covalent linkages between the inhibitor and FAD.
Another problem that complicates the docking method is the size of the dual inhibitors: 440 and
441. These novel dual inhibitors are very bulky to dock using standard docking tools and may
clash with the amino acid residues that surround the binding site.
Due to these problems, it was decided to use Tethered Molecular Docking (TMD) to dock the
inhibitors. This is a software from Dr Ian Haworth’s lab that is used for the simulation of desired
covalent bonds between inhibitors and the enzyme while running different combinations of bond
and torsion angles of the ligands to generate protein-ligand complexes or models.

Figure 2: Structure of the Dual MAO-A/HDAC inhibitors that are analyzed

From previous studies carried out on other MAO-A/HDAC dual inhibitors, certain MAO-A
crystal structures were constructed using MODELLER and an MAO-A crytal structure was
finalized for carrying out docking and solvation free energy calculations. (Thomas Asbell
Masters Thesis, 2021)

5
TMD setup: Generating Clorgyline-Based Inhibitor .zmat files:
In order to use TMD, certain input files must be prepared. zmat files are one of the most
important files for running TMD. This file is made for each inhibitor of interest. These files
contain detailed information on the connectivity of each atom to one another, as well as bond
lengths, angles and torsions of each bond. These files have been generated in a previously
conducted MAO-A/HDAC inhibitor study (Refered from Thomas Asbell MS thesis). The rest of
the inhibitor structure was generated using the correct molecular geometry for each bond, each
.zmat file contains a few differences in order to uniquely store the geometries of the atoms in
each inhibitor.
6
4 C 1 2 3 1.37 86.80 -110.58
5 C 4 1 2 1.36 122.76 -104.53
6 C 5 4 1 1.37 125.30 -145.23
7 N 6 5 4 1.37 118.60 -179.58
8 C 7 6 5 1.47 120.01 00.00
9 C 7 6 8 1.47 119.98 -180.00
10 C 9 7 6 1.53 109.41 68.58
11 C 10 9 7 1.53 109.42 120.00
12 O 11 10 9 1.43 109.42 -52.00
13 C 12 11 10 1.36 106.77 -179.98
14 C 13 12 11 1.39 120.00 179.97
15 Cl 14 13 12 1.74 119.99 0.01
16 C 14 13 15 1.38 119.97 179.95
17 C 16 14 13 1.39 119.95 0.01
18 N 17 16 14 1.40 119.94 179.95
19 C 18 17 16 1.35 119.97 00.00
20 C 19 18 17 1.53 109.40 180.00
21 C 20 19 18 1.40 120.00 180.00
22 C 21 20 19 1.40 120.00 180.00
23 C 22 21 20 1.40 120.00 0.00
24 C 23 22 21 1.40 120.00 0.00
25 C 24 23 22 1.40 120.00 0.00
26 C 23 22 21 1.50 120.00 180.00
27 C 26 23 22 1.37 120.00 180.00
28 C 27 26 23 1.51 120.00 180.00
29 N 28 27 26 1.35 120.05 179.96
30 O 29 28 27 1.42 120.05 -179.98
31 O 28 27 29 1.21 119.97 -179.95
32 C 17 16 18 1.39 120.03 -179.91
33 C 13 12 14 1.39 120.02 179.65
34 H01 8 7 10 1.09 109.46 -4.96
35 H02 9 7 34 1.09 109.44 120.87
36 H03 9 7 34 1.09 109.46 00.00
37 H04 6 5 7 1.08 119.98 179.91
38 H05 5 6 37 1.08 119.97 -174.52
39 H06 4 5 6 1.08 119.94 -0.11
40 H07 8 7 10 1.09 109.47 119.99
41 H08 8 7 10 1.09 109.46 -119.99
42 H09 10 8 11 1.09 109.53 119.94
43 H10 10 8 11 1.09 109.52 -119.97
44 H11 11 10 12 1.09 109.41 119.88
45 H12 11 10 12 1.09 109.48 -119.97
46 H13 16 14 17 1.08 119.96 179.93
47 H14 18 17 19 0.97 119.94 179.84
48 H15 19 18 20 1.09 109.40 119.98
49 H16 19 18 20 1.09 109.51 -120.04
50 H17 21 20 22 1.09 120.00 180.00
51 H18 22 21 23 1.09 120.00 180.00
52 H19 24 23 25 1.09 120.00 180.00
53 H20 25 24 20 1.09 120.00 180.00
54 H21 29 28 30 0.97 119.98 -179.90
55 H22 30 29 28 0.97 106.84 179.98
56 H23 32 17 16 1.08 120.04 179.94
57 H24 33 13 12 1.08 119.94 0.42
58 H05 26 23 22 1.08 119.97 00.00
59 H06 27 26 23 1.08 119.94 00.00

Figure 3: Raw .zmat file of 440 using torsions and bond angles of clorgyline in the crystal
structure (Thomas Asbell Masters Thesis, 2021)
7

4 C 1 2 3 1.37 86.80 -110.58
5 C 4 1 2 1.36 122.76 -104.53
6 C 5 4 1 1.37 125.30 -145.23
7 N 6 5 4 1.37 118.60 -179.58
8 C 7 6 5 1.47 120.01 00.00
9 C 7 6 8 1.47 119.98 -180.00
10 C 9 7 6 1.53 109.41 68.58
11 C 10 9 7 1.53 109.42 120.00
12 O 11 10 9 1.43 109.42 -52.00
13 C 12 11 10 1.36 106.77 -179.98
14 C 13 12 11 1.39 120.00 179.97
15 Cl 14 13 12 1.74 119.99 0.01
16 C 14 13 15 1.38 119.97 179.95
17 C 16 14 13 1.39 119.95 0.01
18 N 17 16 14 1.40 119.94 179.95
19 C 18 17 16 1.35 119.97 00.00
20 C 19 18 17 1.53 109.40 180.00
21 C 20 19 18 1.40 120.00 180.00
22 C 21 20 19 1.40 120.00 180.00
23 C 22 21 20 1.40 120.00 0.00
24 C 23 22 21 1.40 120.00 0.00
25 C 24 23 22 1.40 120.00 0.00
26 C 23 22 21 1.51 120.00 180.00
27 N 26 23 22 1.35 120.05 179.96
28 O 27 26 23 1.42 120.05 -179.98
29 O 26 23 27 1.21 119.97 -179.95
30 C 17 16 18 1.39 120.03 -179.91
31 C 13 12 14 1.39 120.02 179.65
32 H01 8 7 10 1.09 109.46 -4.96
33 H02 9 7 32 1.09 109.44 120.87
34 H03 9 7 32 1.09 109.46 00.00
35 H04 6 5 7 1.08 119.98 179.91
36 H05 5 6 35 1.08 119.97 -174.52
37 H06 4 5 6 1.08 119.94 -0.11
38 H08 8 7 10 1.09 109.47 119.99
49 H09 8 7 10 1.09 109.46 -119.99
40 H09 10 8 11 1.09 109.53 119.94
41 H10 10 8 11 1.09 109.52 -119.97
42 H11 11 10 12 1.09 109.41 119.88
43 H12 11 10 12 1.09 109.48 -119.97
44 H13 16 14 17 1.08 119.96 179.93
45 H14 18 17 19 0.97 119.94 179.84
46 H15 19 18 20 1.09 109.40 119.98
47 H16 19 18 20 1.09 109.51 -120.04
48 H17 21 20 22 1.09 120.00 180.00
49 H18 22 21 23 1.09 120.00 180.00
50 H19 24 23 25 1.09 120.00 180.00
51 H20 25 24 20 1.09 120.00 180.00
52 H21 27 26 28 0.97 119.98 -179.90
53 H22 28 27 26 0.97 106.84 179.98
54 H23 30 17 16 1.08 120.04 179.94
55 H24 31 13 12 1.08 119.94 0.42

Figure 4 : Raw .zmat file of 441 using torsions and bond angles of clorgyline in the crystal
structure.
8
TMD Molecular Docking :
With .zmat files ready, TMD was run using 2BXS protein (crystal structure of human MAO-A
complexed with clorgyline) by using varying different torsions within each .zmat file to explore
possible poses each inhibitor may take while bound to the FAD. The torsions are tested in each
TMD run and the results of the run are shown in Table 1. These runs were carried out in sets of
two for each of the two inhibitors. Changes are made particularly to atom 19 as certain torsion
angles were causing clashes, therefore the generation of two sets. These runs were conducted
using different protein models as protein loops were reconstructed as it was observed that the
dual inhibitors have residues that caused steric clashes at the entrance of the MAO-A binding
cavity. It has been noted in a few studies that the entrance of the binding cavity is too small for
any inhibitor/ substrate to enter.
8
Since these loops are flexible it is possible that they can be
adjusted into different conformations in order to make the inhibitor/substrate enter the active site.
With the help of MODELLER 5 protein structures were generated, these models are then used as
templates for docking these dual inhibitors. Generation of these models were carried out in
previous studies for MAO-A/HDAC dual inhibitors 357 and 359 from which these models were
taken. (Reference to Thomas Asbell masters thesis).
9

Figure 5 : Structures of protein models 1-5 with strands created using MODELLER
(Reference:Thomas Asbell Masters Thesis)
10
Table 1: Torsion angles tested using TMD exploring different possible conformations of 440
and 441 Model 1-5 , set 1, set2
440
Model
Number
Set Atom Number Set torsion to: End with Torsion: In increments of:
1 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5
Angstroms

Maximum Tolerated Clashes 5

Total number of Poses 4

1 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 21

2 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 4

11
2 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 21

3 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 4

3 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 21

4 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
12
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 4

4 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 21

5 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 4

5 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 21

13
441
Model
Number
Set Atom Number Set torsion to: End with Torsion: In increments of:
1 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 10

1 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 6

2 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 12

2 2 1 0 0 0
5 0 0 0
14 185 225 10
14
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 6

3 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 10

3 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 6

4 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

15
Total number of Poses 10

4 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 6

5 1 1 0 0 0
5 0 0 0
14 185 225 10
19 0 180 180
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å
Maximum Tolerated Clashes 5

Total number of Poses 10

5 2 1 0 0 0
5 0 0 0
14 185 225 10
19 160 200 10
20 60 300 120
21 0 330 30
27 0 180 180
Maximum Clash Distance 1.5 Å

Maximum Tolerated Clashes 5

Total number of Poses 6

It may be the case that these poses generated are energetically favourable, but in order to make
an estimation of the energetic favorability of these inhibitor complexes and calculation, a
solvation analysis was conducted.
16
Solvation Analysis:
The results from docking calculations only show the possible poses of the inhibitor that can bind
to MAO-A. In vitro IC50 MAO A inhibitory activities of each dual inhibitor in GL26 cells were
obtained (Table 2 )
5
. Certain dual inhibitors exhibited very potent MAO A inhibition relative to
other inhibitors. However our focus is only on 440 and 441. With differences in affinity for
MAO A, the generated poses should reflect this difference through variations in the energetic
feasibility of each pose. To understand the differences in inhibitor affinities compared to one
another, a solvation analysis of each pose was conducted. Solvation calculations shows us how
many water molecules were displaced from the binding cavity, how many protein-ligand bonds
were broken. The WATGEN software uses an algorithm to use this docking data to compute
estimated free energy.
Compound MAOA Inhibitory IC50
(M)
357 2.64 × 10
-8

359 1.91 × 10
-12

440 4.03 × 10
-9

441 4.09 × 10
-10

Table 2 : Experimental MAO-A IC50 data using GL26 cells
5

Solvation-free energy calculations have their limitations. Since the clorgyline is bound to FAD it
makes it very hydrophobic. This can cause a few solvation issues. It is likely solvation analysis
either understimated or ovestimated certain free energies. One issue could be the fact that the
software assumes all the protein models to be equally possible. Keeping these issues in mind,
solvation free energy data needs to be used in a certain way to predict the affinity values. Also, it
would be important to mention that these are novel dual inhibitors, to predict affinity values
using a model would be ideal and would be specifically designed for these inhibitors making
affinity prediction using a machine learning model a unique and possibly faster way to get
results. This machine learning mdoel was generated using a software called KNIME, as
explained in Chapter 3.
17
Chapter 3: Machine Learning Model
In order to understand solvation and predict the affinity values, a machine learning model was
built using previously published data for more than 9000 protein-ligand complexes.
6
. With
technological advances in high-throughput screening, the designing of various virtual libraries
has now led to the pursuit of finding new and challenging drug targets. Data mining is one such
method that has been vastly used and specifically designed to extract key pieces of information
from large data sets, especially biological datasets. This method has been particularly helpful in
drug discovery. Open source or freely available software has gained momentum in recent times
and has proven to be helpful in academic settings. One such software is KNIME (Konstanz
Information Miner). It contributes to the drug discovery pipeline through a visual assembly of
data workflows that are drawn from a large repository of tools.
4

KNIME uses a “workflow” framework. These “workflows” present an effective way to manage
the scientific information provided as input to help standardize protocols and aid in data analysis.
KNIME allows workflows to be quickly customized which presents a clear visual interface.
KNIME nodes are used to perform a wide array of functions like read/write, various statistical
analysis and machine learning algorithms.
4

Figure 6 : General KNIME workbench
illustrating possible views, this includes workflow editor (central panel), workflow lists (top
left), node repository (left panel), current workflow outline( bottom panel
18

Figure 7 : Complete KNIME (workflow) model used.

This particular model was constructed in KNIME and used to generate 19 different combination
models using certain parameters. The model was first tested and generated based on a dataset
that contains published data of more than 9000 compounds
6
. Results are generated using a
machine learning model called “Gradient Boosted Trees Model “. Machine learning models
primarily work on a concept called “decision trees”, these are models which are used to make
predictions based on how a previous set of questions was answered. Models are trained and
tested based on the data we give as input. Gradient Boosted Trees is a type of decision tree that is
used to predict a target label, by minimizing the overall prediction error. Here the data for more
than 9000 compounds are divided into “training” and “test” sets: 90% of the data is partitioned
into the training set and 10% into the test set. The model will learn from the training set and
apply the workflow nodes mentioned in Figure 7 to the data. Functions of the nodes mention in
Table 3.
In order to understand the workflow, the arrows need to be followed. The workflow consists of
“nodes”. Nodes represent individual tasks. Displayed as colored boxes with an input and output
port, nodes can perform various functions, this includes transforming data, reading/writing files,
training models and so on. Nodes can be added or deleted according to the need of the model,
therefore the numbering of the nodes can vary and may be inconsistent.
19
Nodes Function
Excel Reader Read single and multiple (xlsx, xlm etc.)files
at the same time
String Manipulation Manipulates the string, can change the format
of the input data which suits the model for
better readability
Column Filter Table is filtered, which coloumns (categories)
can be included or excluded.
Row Filter Row from table can be filtered according to
certain criteria, rows with certain IDs and
categories can be selected.
Partitioning Divides data, into training and test data
Gradient Boosted Trees Learner Generates possible models, input data to learn
from, must contain target column
Gradient Boosted Trees Predictor Output predicted data
Joiner Joins two tables in a database, similar named
columns can be combined too.
Column Resorter Columns can be rearranged
Scatter Plots Shows 2D plots
CSV Writer Writes out the the input data table into a file
(excel)
Table 3 : Nodes used and their functions
Based on the workflow, categories were selected and added one by one into the model, different
permutation combinations were used to create a model which shows good results. 19 such
models were made. The categories are important physicochemical properties that are useful in
selecting good drug targets. The prediction affinities are affected based on these categories.
Categories Function
Protein Ligand HB Distance measured from ligand and protein
heavy atoms to water oxygens( Hydrogen
bond interaction)
10

Protein Ligand HF Distance measured from ligand and protein
heavy atoms to water oxygens( Hydrophobic
interaction)
10

Rotatable bonds
Single non ring, non hydrogen atom
10

Molecular weight
Mass of a molecule
10

LogP
Partition coefficient
10

Actual Single Water Bridge in position to form a hydrogen bond with a
protein atom and a ligand atom
10

20
Contact Single Water Bridge 1.0−2.5 Å from the ligand, not matched in
position to form a hydrogen bond with a
protein atom and a ligand atom
10

Absolute Displaced Waters
<1.0 Å from any ligand atom
10

Total New(moved) Ligand Waters Displaced water molecules after ligand binds
to the binding site
10

Total Matched Prot Waters Remaining water molecules that are near the
protein
10

Total Matched Ligand Waters Remaining water molecules that are near the
ligand
10

Total Indirect Displaced Moved Prot Waters Water molecules displaced near biniding sites
that are unaccounted for.
10

Table 4 : Functions of categories used for the model.

With information about categories and a set workflow, the model can now be run for prediction.
Input data given to the model contains 9285 fully solvated structures from MOAD
6
.
1. Based on the arrangement of the nodes in the workflow and after exceuting the final node
for the generation of a CSV file the predicted value .csv excel sheet is generated based on
the categories selected from the column filter.
2. The predicted values from the newly generated .csv is is taken and added to newsheet
with a column containing experimentally predicted values and the new predicted affinity
values.
3. A matrix is now created with this information Figure 8-26 (Model 1-19) to find out the
percentage of correctly predicted values which will help determine the accuracy of the
model depending on each category chosen the prediction of the model varies and
definitely contributes to the accuracy of the model. Data is binned in 1kcal/mol
4. Along with this matrix a visual representation of this data os also needed. Graphs within
Figure 1-19 shows the count of predicted ΔG versus prediction ΔG, every data point in
the graph contains every complex which has an experimental ΔG of the binned 1kcal/mol
values.
5. In Models 1-19, horizontal row from -2 to -14 experimental free energy values, and
vertical columns from -1 to -16 are the values whose count needs to be predicted.
21
6. Data within the matrix is the count of every experimental value that is correctly predicted
to be the experimental value.
7. Bins from -7 to -11 are shown in the graphs as they contained majority of the data and
can be easily visualised.
Model
No
HB HF RB MW ASWB CSWB ADW TNLW TMPW TMLW TIMW LogP
1 X X

2 X X X

3 X X X X

4 X X X X X

5 X X X X X X

6 X X X X X X X

7 X X X X X X X X

8 X X X

X X X X

X
9 X X X X X X X X

X
10 X X X

X X X X

11 X X X X X X

X

X
12 X X X X X X

X

13 X X X

X
14 X X X X

X
15 X X X X X X X X X

X
16 X X X X X X X X X X

X
17 X X X X X X X X X X X X
18 X X X X X X X X

X X X
19 X X X X X X X X X X X
Table 5 : Combinations of 19 different models
Model No: Model number
HB: Protein-Ligand Hydrogen Bond
HF: Protein-Ligand Hydrophobic interactions
RB: Rotatable Bonds
MW: Molecular Weight
ASWB: Actual Single Water Bridges
CSWB: Contact Single Water Bridges
ADW: Absolute Displaced Waters
TNLW: Total New (moved) Lig Waters
TMPW: Total Matched Prot Waters
TMLW: Total Matched Lig Waters
TIMW: Total Indirect Displaced Moved Prot Waters
Log P: Partition Coefficient
22
Model 1:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 0 0 0 0 0 0
-2 0 0 0 0 0 0 0 0 0 0 0 0 0
-3 0 0 0 0 0 0 0 0 0 0 0 0 0
-4 0 0 0 0 0 0 0 0 0 0 0 0 0
-5 0 0 2 5 2 2 4 1 3 1 1 0 0
-6 0 1 1 2 2 3 7 5 2 0 1 0 0
-7 3 7 7 11 21 21 26 14 14 15 4 2 0
-8 0 3 6 3 8 10 7 7 8 6 3 1 1
-9 1 10 11 17 14 24 40 41 29 28 12 8 2
-10 1 1 4 8 16 21 34 24 30 31 26 14 3
-11 2 5 2 9 7 13 18 30 43 29 15 7 2
-12 0 0 1 1 1 4 3 6 2 2 4 0 0
-13 0 0 0 0 0 0 0 0 2 0 0 0 0
-14 0 0 0 0 0 0 0 0 0 0 0 0 0
-15 0 0 0 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 8 : Only HB and HF
used this model was used to check the model, as seen in the graph it does not show good results.

23
Model 2:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 1 0 0 0 0 0
-2 0 1 1 2 1 1 0 0 0 0 0 1 0
-3 0 0 0 0 0 0 0 0 0 0 0 0 0
-4 1 1 2 3 5 5 5 0 2 3 3 0 0
-5 0 4 4 6 3 8 9 5 4 5 3 1 0
-6 2 6 7 14 17 9 15 12 7 7 3 4 0
-7 1 4 6 5 8 13 13 9 6 4 2 2 0
-8 0 2 0 2 7 4 8 9 13 5 4 0 0
-9 1 3 6 11 9 29 24 34 22 19 16 7 2
-10 0 0 3 4 13 21 33 25 29 30 18 9 2
-11 2 6 2 8 8 5 21 29 41 30 11 5 3
-12 0 0 1 1 0 1 7 3 8 6 4 1 1
-13 0 0 2 0 0 1 2 1 1 3 2 2 0
-14 0 0 0 0 0 0 1 0 0 0 0 0 0
-15 0 0 0 0 0 0 1 0 0 0 0 0 0
-16 0 0 0 0 0 1 0 0 0 0 0 0 0

Figure 9: Addition of RB bonds
along with HB, HF shows a slight change, the -9 bin shows a good spike but is very inconsistent.
24
Model 3:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 0 0 0 0 0 0
-2 0 3 0 2 1 0 0 0 0 0 0 0 0
-3 0 0 3 1 4 3 1 2 0 0 0 0 0
-4 2 6 4 2 6 1 6 7 0 0 0 0 0
-5 0 4 8 10 7 9 9 3 2 1 2 0 0
-6 5 2 6 15 13 12 10 6 4 1 0 3 0
-7 0 2 4 5 5 17 22 10 7 6 3 0 0
-8 0 2 1 4 9 10 14 7 11 8 6 2 0
-9 0 4 3 9 8 23 32 31 23 17 10 5 1
-10 0 2 1 3 10 11 20 31 37 32 15 6 0
-11 0 1 2 4 5 10 22 28 38 36 18 5 6
-12 0 1 2 1 3 2 2 1 8 8 8 6 0
-13 0 0 0 0 0 0 1 0 2 1 2 4 1
-14 0 0 0 0 0 0 0 1 0 2 1 0 0
-15 0 0 0 0 0 0 0 0 1 0 1 0 0
-16 0 0 0 0 0 0 0 1 0 0 0 1 0

Figure 10 : Addition of MW
shows improvement in the peaks, specially in in the -9, -10, -11 bins
25
Model 4:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 0 1 0 0 0 0
-2 1 3 0 3 2 1 0 0 0 0 1 0 0
-3 0 2 2 0 2 3 1 2 0 0 1 0 0
-4 0 5 4 3 5 2 3 4 1 1 0 0 0
-5 0 6 9 10 8 7 4 4 1 1 1 0 0
-6 4 5 6 13 18 12 15 7 2 2 0 2 0
-7 1 3 6 8 7 17 17 18 6 4 5 1 0
-8 0 1 3 4 5 12 10 9 8 6 3 2 0
-9 0 2 0 5 13 18 38 32 20 18 11 4 1
-10 0 0 1 1 3 13 18 20 38 27 13 8 0
-11 1 0 2 5 5 10 27 25 43 42 23 5 5
-12 0 0 1 4 2 3 4 3 9 9 6 7 2
-13 0 0 0 0 1 0 1 1 4 1 1 2 0
-14 0 0 0 0 0 0 0 1 0 1 1 0 0
-15 0 0 0 0 0 0 0 1 0 0 0 0 0
-16 0 0 0 0 0 0 1 1 0 0 0 1 0

Figure 11 : Addition of ASWB
shows a good peak for -9, -11, this one of the solvation parameters.This shows improvement in
the peaks and requirement of solvation parameters.
26
Model 5:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 0 1 0 0 0 0
-2 0 5 0 2 2 1 0 0 0 0 0 0 0
-3 0 2 2 0 1 2 0 2 0 0 0 0 0
-4 1 3 6 2 7 1 3 4 2 0 0 0 0
-5 1 5 8 10 6 8 5 4 1 2 1 0 0
-6 4 1 4 18 10 7 12 3 1 2 2 2 1
-7 1 3 8 4 7 24 21 17 7 5 6 0 0
-8 0 3 2 3 11 13 13 15 13 8 3 2 0
-9 0 5 0 6 14 17 32 27 18 21 10 5 1
-10 0 0 1 3 3 16 22 30 33 22 12 5 1
-11 0 0 2 6 7 7 22 23 44 43 22 7 4
-12 0 0 1 2 2 2 7 1 9 7 7 5 1
-13 0 0 0 0 1 0 1 0 4 2 2 5 0
-14 0 0 0 0 0 0 0 2 0 0 1 0 0
-15 0 0 0 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 1 0 0 0 0 1 0

Figure 12 : Addition of CSWB
another solvation parameter, peaks are distinct, specially on the -11 bin.
27
Model 6 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 1 0 0 0 0 0
-2 1 5 0 2 1 0 0 0 0 0 0 0 0
-3 0 2 2 1 2 2 1 1 0 0 0 0 0
-4 1 3 7 1 5 2 3 3 0 0 0 0 0
-5 0 5 9 10 9 7 4 4 4 2 2 0 0
-6 3 2 3 16 14 8 12 7 3 0 3 2 0
-7 2 4 6 8 4 17 26 17 6 5 4 0 0
-8 0 3 3 4 11 16 11 12 8 5 2 3 1
-9 0 3 0 6 13 21 26 25 23 23 12 5 1
-10 0 0 2 2 3 15 26 35 33 25 10 5 1
-11 0 0 1 4 6 9 25 18 39 41 23 4 4
-12 0 0 1 2 2 1 4 4 11 7 8 8 1
-13 0 0 0 0 1 0 0 0 6 3 1 4 0
-14 0 0 0 0 0 0 0 1 0 0 1 0 0
-15 0 0 0 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 1 0 0 1 0 1 0

Figure 13: Addition of ADW
reduction in the kinks is observed, -11 shows a good peak, ADW is another solvation parameter
28
Model 7 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -
14
-1 0 0 0 0 0 0 0 0 0 0 0 0 0
-2 0 2 0 1 0 1 1 0 0 0 0 1 0
-3 0 2 3 0 3 4 1 2 0 0 1 0 0
-4 1 5 4 0 5 2 3 4 0 0 0 1 0
-5 0 6 10 12 4 8 5 4 2 1 1 0 0
-6 3 2 6 15 13 12 11 9 4 2 1 2 1
-7 3 6 4 9 10 19 21 16 7 3 6 2 0
-8 0 1 4 4 9 11 13 7 9 2 1 2 0
-9 0 3 0 10 15 17 27 28 24 24 11 4 0
-10 0 0 1 1 4 17 27 29 34 24 10 6 1
-11 0 0 1 3 6 7 21 21 40 50 23 5 5
-12 0 0 1 0 1 0 6 6 10 5 11 5 1
-13 0 0 0 1 1 0 1 0 3 1 1 3 0
-14 0 0 0 0 0 0 1 1 0 0 0 1 0
-15 0 0 0 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 1 1 0 0 0 0 0

Figure 14 : Addition of TNLW
-10, -11 shows a good peak, TNLW is another solvation parameter
29
Model 8:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 2 0 1 0 0 0 0 0 0 0 0 0
-3 0 1 1 1 1 2 0 0 0 0 0 0 0
-4 1 3 4 3 6 2 4 7 1 0 0 1 0
-5 0 5 4 8 7 6 7 5 5 2 0 1 0
-6 2 5 10 14 10 8 19 7 5 1 1 0 0
-7 3 5 4 5 13 17 20 15 9 8 2 2 0
-8 0 1 5 4 5 9 9 13 7 6 3 0 0
-9 1 1 1 9 9 23 20 29 27 23 16 5 2
-10 0 1 3 4 10 20 33 27 34 26 20 5 2
-11 0 0 0 6 8 8 21 23 38 32 14 10 3
-12 0 1 2 0 0 3 2 1 7 12 9 5 0
-13 0 0 0 0 2 0 3 1 0 2 1 2 1
-14 0 1 0 1 0 0 1 0 0 0 0 1 0
-15 0 0 0 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 15: Removal of MW
Replacing it with LogP, messes with the peaks and it can be observed that MW may be an
important parameter. However, due to the solvation parameters the data peaks well at -9, -10, -11
30
Model 9 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 2 0 0 0 0 0 0 0 0 0 0 0
-2 1 4 0 1 0 0 0 1 0 0 0 0 0
-3 0 0 2 1 2 3 0 0 0 0 0 0 0
-4 1 1 1 2 4 2 5 3 1 0 0 1 0
-5 0 6 14 8 7 5 6 7 3 0 2 0 0
-6 3 5 6 15 21 12 9 7 5 3 2 1 0
-7 2 5 2 6 6 23 16 19 7 4 4 0 0
-8 0 1 3 9 5 10 14 6 12 3 2 0 0
-9 0 3 1 5 12 21 36 29 24 19 11 5 1
-10 0 0 1 4 4 12 23 28 26 19 8 5 0
-11 0 0 1 3 8 9 23 21 46 51 25 9 5
-12 0 0 2 2 1 1 6 3 7 8 7 4 1
-13 0 0 0 0 1 0 1 0 2 5 2 6 1
-14 0 0 0 0 0 0 0 2 0 0 1 0 0
-15 0 0 1 0 0 0 0 0 0 0 2 0 0
-16 0 0 0 0 0 0 0 2 0 0 0 1 0

Figure 16: Re-addition of MW
along with HB, HF, MW, RB, ASWB, CSWB, ADW, TNMLW shows an excellent peak at -9,-
11, showing that addition MW seems to be improving the model.
31
Model 10 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 2 0 1 0 0 0 0 0 0 0 0 0
-3 0 1 1 1 1 2 0 0 0 0 0 0 0
-4 1 3 4 3 6 2 4 7 1 0 0 1 0
-5 0 5 4 8 7 6 7 5 5 2 0 1 0
-6 2 5 10 14 10 8 19 7 5 1 1 0 0
-7 3 5 4 5 13 17 20 15 9 8 2 2 0
-8 0 1 5 4 5 9 9 13 7 6 3 0 0
-9 1 1 1 9 9 23 20 29 27 23 16 5 2
-10 0 1 3 4 10 20 33 27 34 26 20 5 2
-11 0 0 0 6 8 8 21 23 38 32 14 10 3
-12 0 1 2 0 0 3 2 1 7 12 9 5 0
-13 0 0 0 0 2 0 3 1 0 2 1 2 1
-14 0 1 0 1 0 0 1 0 0 0 0 1 0
-15 0 0 0 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 17: Removal of both MW and LogP
shows interesting observation of an increase in kinks, however due to the solvation parameters
being intact -9, -11 continue to show good peaks.
32
Model 11 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 2 0 1 2 0 1 0 1 0 0 0 0
-3 0 1 2 1 2 4 1 2 0 0 0 0 0
-4 1 4 6 0 6 1 3 3 1 0 0 1 0
-5 0 8 10 11 4 9 5 7 2 1 0 0 0
-6 3 2 6 18 11 11 7 5 5 3 2 3 0
-7 2 6 4 7 8 21 20 16 5 4 6 1 0
-8 0 0 2 5 10 10 16 8 9 4 4 2 0
-9 0 3 1 6 17 15 32 26 22 28 14 4 1
-10 0 0 0 2 1 16 17 34 39 28 12 6 0
-11 1 0 1 3 8 8 28 20 37 38 21 7 5
-12 0 0 1 0 1 1 4 3 8 4 6 3 1
-13 0 0 0 1 1 1 1 2 4 2 1 5 1
-14 0 0 0 1 0 1 2 1 0 0 0 0 0
-15 0 0 1 0 0 0 1 0 0 0 0 0 0
-16 0 0 0 0 0 0 1 1 0 0 0 0 0

Figure 18 : Removal of ADW
one of the solvation parameters, shows an increase in the kinks and very irregular patterns, -10, -
11 still show a decent result.
33
Model 12 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 2 0 1 2 0 1 0 1 0 0 0 0
-3 0 1 2 1 2 4 1 2 0 0 0 0 0
-4 1 4 6 0 6 1 3 3 1 0 0 1 0
-5 0 8 10 11 4 9 5 7 2 1 0 0 0
-6 3 2 6 18 11 11 7 5 5 3 2 3 0
-7 2 6 4 7 8 21 20 16 5 4 6 1 0
-8 0 0 2 5 10 10 16 8 9 4 4 2 0
-9 0 3 1 6 17 15 32 26 22 28 14 4 1
-10 0 0 0 2 1 16 17 34 39 28 12 6 0
-11 1 0 1 3 8 8 28 20 37 38 21 7 5
-12 0 0 1 0 1 1 4 3 8 4 6 3 1
-13 0 0 0 1 1 1 1 2 4 2 1 5 1
-14 0 0 0 1 0 1 2 1 0 0 0 0 0
-15 0 0 1 0 0 0 1 0 0 0 0 0 0
-16 0 0 0 0 0 0 1 1 0 0 0 0 0

Figure 19: Removal of Log P and ADW
not much difference seen compared to Figure 11, shows that LogP may not be contributing to the
model.
34
Model 13 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 4 0 3 1 0 0 1 0 0 0 0 0
-3 0 2 2 0 3 1 0 0 0 0 0 0 0
-4 2 2 2 5 4 1 2 2 3 1 0 0 0
-5 0 7 8 6 7 6 7 3 3 3 1 0 0
-6 3 2 7 6 17 11 10 13 8 3 5 2 0
-7 1 3 6 14 11 22 25 10 10 9 5 0 0
-8 0 3 3 2 4 16 18 12 11 6 4 1 1
-9 0 1 3 8 12 14 33 29 27 23 10 6 0
-10 1 0 0 5 6 11 17 28 26 23 14 8 1
-11 0 2 1 4 5 13 21 20 38 39 16 7 4
-12 0 0 2 3 1 1 4 7 5 4 8 4 2
-13 0 0 0 0 0 2 2 1 2 1 2 3 0
-14 0 0 0 0 0 0 0 0 0 0 0 0 0
-15 0 0 0 0 0 0 0 1 0 0 1 0 0
-16 0 0 0 0 0 0 0 1 0 0 0 1 0

Figure 20: Removal of all solvation categories except LogP, HB,HF.
Shows inconsistency in peaks Solvation does seem to be of importance.
35
Model 14 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 5 0 2 1 0 0 1 0 0 0 0 0
-3 0 0 2 1 5 1 0 1 0 0 0 0 0
-4 3 5 2 1 4 1 4 5 0 0 1 0 0
-5 1 4 13 8 6 8 6 4 2 1 2 0 0
-6 1 3 6 13 18 14 9 7 5 2 0 2 0
-7 2 4 3 9 8 22 24 12 7 4 3 0 0
-8 0 3 3 6 4 12 15 12 9 8 1 1 0
-9 0 2 1 7 9 21 35 29 31 20 8 4 1
-10 0 0 1 3 8 9 18 36 24 24 17 10 0
-11 0 0 2 3 7 8 21 16 44 44 22 4 5
-12 0 0 1 2 1 1 5 2 7 7 8 3 1
-13 0 0 0 0 0 0 1 0 3 1 2 7 1
-14 0 0 0 1 0 1 0 1 1 1 1 0 0
-15 0 0 0 0 0 0 0 1 0 0 1 0 0
-16 0 0 0 0 0 0 1 1 0 0 0 1 0

Figure 21: Removal of solvation parameters(ASWB, CSWB, ADW, TNMLW)
readdition of MW, along with LogP, HB,HF. Increse in -11 bin, model is slighlty improved due
to addition of MW
36
Model 15 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 3 1 0 0 0 1 0 0 0 0 0 0
-3 0 0 2 3 4 3 0 0 0 0 0 0 0
-4 3 4 3 3 4 2 3 1 1 0 0 1 0
-5 0 5 11 9 6 6 6 6 1 0 2 0 0
-6 3 4 5 14 23 11 8 8 7 3 2 0 0
-7 1 6 5 9 5 22 23 20 7 4 3 0 0
-8 0 1 3 5 4 9 14 12 8 2 1 1 0
-9 0 3 0 5 11 19 36 29 23 20 12 6 1
-10 0 0 1 5 8 11 20 29 31 23 8 5 1
-11 0 0 1 2 4 11 21 20 40 52 30 8 5
-12 0 0 1 1 1 3 7 2 9 6 6 6 0
-13 0 0 0 0 1 0 0 0 4 2 1 5 1
-14 0 0 0 0 0 1 0 1 2 0 1 0 0
-15 0 0 1 0 0 0 0 0 0 0 0 0 0
-16 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 22: Readdition of solvation parameters, along with MW, logP and TMPW.
This definitely improves the model, especially -9, -11 bin, the addition of new solvation
parameter TMPW improves model.
37
Model 16 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 0 0 0 0 0 0
-2 0 3 1 1 1 1 0 1 0 0 0 0 0
-3 0 0 2 1 3 2 0 0 0 0 0 0 0
-4 3 5 2 4 5 2 5 2 1 1 0 1 0
-5 0 5 13 9 4 4 5 7 1 0 1 0 0
-6 3 3 5 17 19 12 8 5 5 2 4 0 0
-7 1 6 3 7 10 22 21 16 8 6 1 2 0
-8 0 2 2 3 7 14 14 14 10 2 5 0 1
-9 0 2 1 6 12 17 33 25 22 19 8 4 1
-10 0 0 0 5 3 14 24 29 31 26 10 5 0
-11 0 1 1 2 6 6 20 21 44 47 28 11 4
-12 0 0 2 1 1 3 7 6 5 5 6 6 1
-13 0 0 0 0 0 1 2 0 3 4 2 2 1
-14 0 0 0 0 0 0 0 1 2 0 1 1 0
-15 0 0 1 0 0 0 0 0 1 0 0 0 0
-16 0 0 1 0 0 0 0 1 0 0 0 0 0

Figure 23: The addition of another solvation parameter TMLW
shows a good separation in the peaks, especially -11 bin, moel shows the more we add solvation
parameters the better the model gets, however there is not too much difference observed from
Figure 22.
38
Model 17 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 0 0 0 0 0 0 0 0
-2 0 2 0 1 2 1 0 1 1 0 0 1 0
-3 0 0 2 3 3 4 0 1 0 2 0 1 0
-4 3 3 5 1 2 2 4 2 1 0 0 0 0
-5 0 8 10 8 4 5 3 4 2 0 1 0 0
-6 3 4 5 15 18 10 6 8 6 5 4 0 0
-7 1 4 2 12 13 23 26 19 7 5 2 0 0
-8 0 2 3 4 6 11 20 8 10 2 4 1 0
-9 0 2 1 5 11 21 32 35 21 23 11 7 1
-10 0 1 1 3 5 14 22 26 28 20 12 5 0
-11 0 1 2 3 7 6 20 19 45 46 22 9 4
-12 0 0 0 1 0 1 4 3 9 6 7 5 2
-13 0 0 2 0 0 0 2 0 1 1 2 3 1
-14 0 0 0 0 0 0 0 2 2 2 0 0 0
-15 0 0 1 0 0 0 0 0 0 0 1 0 0
-16 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 24 : Addition of TIMW
another solvation parameter. Does not show too much difference but the model is able to show
better separation and decent results.
39
Model 18:

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 1 0 0 0 0 0 0 0 0 0 0 0
-2 0 1 0 1 2 1 1 1 0 0 0 1 0
-3 0 1 2 3 2 2 0 1 0 0 0 0 0
-4 3 2 2 2 4 3 5 1 0 1 1 1 0
-5 0 9 12 7 4 4 4 5 0 0 1 1 0
-6 2 5 6 18 21 14 6 9 5 2 1 0 0
-7 2 4 4 6 9 21 23 18 7 5 3 0 0
-8 0 1 4 3 6 14 13 10 7 4 5 2 0
-9 0 1 0 7 11 21 33 31 23 22 15 5 1
-10 0 1 0 6 3 9 26 33 35 29 8 5 1
-11 0 1 2 2 7 7 21 15 45 40 24 8 4
-12 0 0 1 1 1 1 6 3 7 6 5 6 1
-13 0 0 0 0 1 0 1 0 2 3 1 3 1
-14 0 0 0 0 0 1 0 0 2 0 1 0 0
-15 0 0 1 0 0 0 0 0 0 0 1 0 0
-16 0 0 0 0 0 0 0 1 0 0 0 0 0

Figure 25: Removal of TMPW
separation reduces and seems to be an increase in the kinks, it shows that TMPW may be an
important category and may affect the prediction of the model.
40
Model 19 :

-2 -3 -4 -5 -6 -7 -8 -9 -10 -11 -12 -13 -14
-1 0 0 0 0 0 1 0 0 0 0 0 0 0
-2 1 2 0 1 1 1 1 0 1 0 0 0 0
-3 0 0 2 2 2 3 0 1 1 0 0 0 0
-4 0 2 4 4 4 3 2 3 1 0 1 1 0
-5 0 9 6 13 6 9 6 5 2 1 1 0 0
-6 3 3 8 13 14 12 11 7 2 2 1 1 0
-7 2 6 3 7 6 18 25 15 4 4 2 1 0
-8 1 2 5 3 11 12 18 10 13 2 5 3 0
-9 0 2 2 7 12 14 23 29 24 23 11 6 1
-10 0 1 2 2 6 14 24 27 36 30 11 5 0
-11 0 0 0 3 7 10 23 24 37 44 23 6 6
-12 0 0 1 1 1 1 5 3 7 4 10 4 1
-13 0 0 0 0 1 0 1 1 2 2 1 4 0
-14 0 0 0 0 0 0 0 0 3 0 0 1 0
-15 0 0 1 0 0 0 0 1 0 0 0 0 0
-16 0 0 0 0 0 0 0 2 0 0 0 0 0

Figure 26: Removal of LogP and readdition of TMPW
better separation is observed, reduction in kinks, model shows the importance of TMPW and
LogP and definitely improves the model.
41
1 0.0% 0.0% 5.9% 12.5% 35.2% 34.7% 52.5% 56.3% 76.7% 55.4% 28.8% 0.0% 0.0%
2 0.0% 7.4% 17.6% 41.1% 39.4% 26.5% 32.4% 53.1% 69.2% 58.9% 25.8% 9.4% 0.0%
3 0.0% 33.3% 44.1% 48.2% 35.2% 39.8% 48.9% 53.9% 73.7% 67.9% 42.4% 31.3% 12.5%
4 14.3% 37.0% 44.1% 46.4% 46.5% 41.8% 46.8% 47.7% 75.9% 69.6% 45.5% 28.1% 0.0%
5 0.0% 37.0% 47.1% 53.6% 32.4% 44.9% 47.5% 56.3% 71.4% 64.3% 47.0% 31.3% 0.0%
6 14.3% 37.0% 52.9% 48.2% 38.0% 41.8% 45.3% 56.3% 71.4% 65.2% 48.5% 37.5% 0.0%
7 0.0% 33.3% 50.0% 48.2% 38.0% 42.9% 43.9% 50.0% 73.7% 70.5% 53.0% 28.1% 0.0%
8 0.0% 22.2% 26.5% 44.6% 42.3% 34.7% 35.3% 53.9% 74.4% 62.5% 36.4% 25.0% 12.5%
9 14.3% 18.5% 50.0% 44.6% 47.9% 45.9% 47.5% 49.2% 72.2% 69.6% 51.5% 31.3% 12.5%
10 0.0% 22.2% 26.5% 44.6% 42.3% 34.7% 35.3% 53.9% 74.4% 62.5% 36.4% 25.0% 12.5%
11 0.0% 25.9% 52.9% 51.8% 32.4% 42.9% 48.9% 53.1% 73.7% 62.5% 42.4% 25.0% 12.5%
12 0.0% 25.9% 52.9% 51.8% 32.4% 42.9% 48.9% 53.1% 73.7% 62.5% 42.4% 25.0% 12.5%
13 0.0% 29.6% 35.3% 30.4% 49.3% 50.0% 54.7% 53.9% 68.4% 58.9% 39.4% 21.9% 0.0%
14 0.0% 37.0% 50.0% 39.3% 45.1% 49.0% 53.2% 60.2% 74.4% 67.0% 48.5% 31.3% 12.5%
15 0.0% 25.9% 47.1% 46.4% 47.9% 42.9% 52.5% 54.7% 70.7% 72.3% 56.1% 34.4% 12.5%
16 0.0% 29.6% 50.0% 53.6% 46.5% 49.0% 48.9% 53.1% 72.9% 69.6% 54.5% 28.1% 12.5%
17 0.0% 18.5% 50.0% 42.9% 49.3% 44.9% 56.1% 53.9% 70.7% 64.3% 47.0% 25.0% 12.5%
18 0.0% 14.8% 47.1% 48.2% 47.9% 50.0% 49.6% 57.8% 77.4% 67.0% 45.5% 28.1% 12.5%
19 14.3% 14.8% 35.3% 53.6% 36.6% 42.9% 47.5% 51.6% 72.9% 69.6% 51.5% 28.1% 0.0%

Figure 27 : Shows the percentage prediction or accuracy for each of the models.
This shows the accuracy for each DeltaG value.
It can be observed that Model 15 gives a better prediction comapred to the rest, based on the
prediction accuracy for -10, -11 bins. It means the model has correctly predicted 70.7% -10
values and 72.3% -11 values respectively. The other models ie Models 16, 17, 18, 19 show good
results as well. This proves the importance of solvation and possibly shows a more accurate way
of calculating solvation free energy.
42
Chapter 4 : Application of model
With model being finalised, it needs to be applied to the dual inhibitors 440, 441. Model 15 will
now be used to predict affinity values for the dual inhibitors.

Figure 28 : Model 15 used for MOA-A/HDAC dual inhibitor affinity prediction.
The difference in this workflow is that the partitioning will be different. Since the model has
been trained on >9000 compound data, all the 9,218 compounds will be used as the training data
set and the test set would be empty since the model would make its prediction.
The prediction yielded 183 possible results. Out of which the average affinity values for 440 is
calculated to be -9.32 and 441 is calculated to be -8.97. The model predicted certain poses to be
energetically more favorable Figures 29 and 30.

Figure 29 : Energetically most favourable position calculated for 440 was from set 2
43

Figure 30 : Energetically most favourable position calculated for 441 was from set 2
HDAC and MAO A dual inhibitors were docked using TMD in this project. Each result from the
TMD docking was then subjected to a solvation analysis. However, these are novel inhibitors
and need more knowledge of the structure. TMD itself was quite crucial in the docking process.
The program provided a novel way to model irreversible inhibitors docked to a specific cofactor
in the protein, a feat unachievable using conventional and publicly available software due to the
difficulties in modelling the inhibitor complex and issues recognizing cofactors within the
program itself. While TMD cannot provide information on the most energetically favorable
poses in each docking attempt, TMD coupled with the solvation analysis data can be possibly
used to do so. To make this process easier the inhibitory effect needs to be predicted and a
machine learning model makes the process less challenging.
This study, therefore, provides a method by which future attempts to dock an irreversible
inhibitor to a protein can prove to be successful. These are preliminary results and need to be
worked on further. The model can be more refined with better category combinations. Future
attempts to design MAO-A inhibitors (or dual inhibitors) should take note of this and consider
the length of the inhibitor as well as the placement of hydrophilic residues to take advantage of
44
this tendency of hydrophilic residues that interact with the bulk water at the entrance of the
binding cavity.
45
References
1. Gaweska, H., & Fitzpatrick, P. F. (2011). Structures and mechanism of the
monoamine oxidase family. BioMolecular Concepts, 2(5), 365–377.

https://doi.org/10.1515/bmc.2011.030

2. Kushal, S., Wang, W., Vaikari, V. P., Kota, R., Chen, K., Yeh, T. S., Jhaveri, N., Groshen, S. L.,
Olenyuk, B. Z., Chen, T. C., Hofman, F. M., & Shih, J. C. (2016). Monoamine oxidase A (MAO A)
inhibitors decrease glioma progression. Oncotarget, 7(12), 13842 –13853.

https://doi.org/10.18632/oncotarget.7283

3. Li, P. C., Chen, S. Y., Xiangfei, D., Mao, C., Wu, C. H., & Shih, J. C. (2020). PAMs inhibits
monoamine oxidase a activity and reduces glioma tumor growth, a potential adjuvant treatment
for glioma. BMC complementary medicine and therapies, 20(1), 252.

4. Mazanetz, M. P., Marmon, R. J., Reisser, C. B., & Morao, I. (2012). Drug discovery applications for
KNIME: an open source data mining platform. Current topics in medicinal chemistry, 12(18),
1965 –1979.

https://doi.org/10.2174/156802612804910331

5. Mehndiratta, S., Qian, B., Chuang, J. Y., Liou, J. P., & Shih, J. C. (2022). N-Methylpropargylamine-
Conjugated Hydroxamic Acids as Dual Inhibitors of Monoamine Oxidase A and Histone
Deacetylase for Glioma Treatment. Journal of medicinal chemistry, 65(3), 2208 –2224.

https://doi.org/10.1021/acs.jmedchem.1c01726

6. Morningstar-Kywi, N., Wang, K., Asbell, T. R., Wang, Z., Giles, J. B., Lai, J., Brill, D., Sutch, B. T., &
Haworth, I. S. (2022). Prediction of Water Distributions and Displacement at Protein-Ligand
Interfaces. Journal of chemical information and modeling, 62(6), 1489 –1497.

https://doi.org/10.1021/acs.jcim.1c01266

7. Seto, E., & Yoshida, M. (2014). Erasers of histone acetylation: the histone deacetylase enzymes.
Cold Spring Harbor perspectives in biology, 6(4), a018713.

https://doi.org/10.1101/cshperspect.a018713

46
8. Son, S. Y., Ma, J., Kondou, Y., Yoshimura, M., Yamashita, E., & Tsukihara, T. (2008). Structure of
human monoamine oxidase A at 2.2-A resolution: the control of opening the entry for
substrates/inhibitors. Proceedings of the National Academy of Sciences of the United States of
America, 105(15), 5739 –5744.

https://doi.org/10.1073/pnas.0710626105

9. Wang, K., Luo, J., Yeh, S., You, B., Meng, J., Chang, P., Niu, Y., Li, G., Lu, C., Zhu, Y., Antonarakis, E.
S., Luo, J., Huang, C. P., Xu, W., & Chang, C. (2020). The MAO inhibitors phenelzine and clorgyline
revert enzalutamide resistance in castration resistant prostate cancer. Nature communications,
11(1), 2689.

https://doi.org/10.1038/s41467-020-15396-5

10. Wu, J. B., Shao, C., Li, X., Li, Q., Hu, P., Shi, C., Li, Y., Chen, Y. T., Yin, F., Liao, C. P., Stiles, B. L.,
Zhau, H. E., Shih, J. C., & Chung, L. W. (2014). Monoamine oxidase A mediates prostate
tumorigenesis and cancer metastasis. The Journal of clinical investigation, 124(7), 2891 –2908.

https://doi.org/10.1172/JCI70982

Abstract (if available)

Abstract Molecular docking is one of the most common and established methods in drug discovery. The purpose is to predict and observe the interaction between small molecules and a protein at an atomic level. This method is especially used for identification of novel molecules that may be therapeutically important. However, the most molecular docking software is catered to generating poses for reversible inhibitors. Generating poses or models for irreversible inhibitors is less common.
The aim of this study is to model dual MAO-A/HDAC inhibitors that bind irreversibly to the FAD cofactor in the MAO-A enzyme. In order to generate models, we used an in-house software, TMD (Tethered Molecular Docking), that scanned potentially possible poses from the library of inhibitors that was given as an input. To understand the energetic contributions of these poses, solvation calculations were performed from another in-house software, Watgen5. It was inferred from the poses of these complexes that the effects of solvation in the MOA-A binding site are important in order to drive the ligand binding.
To obtain a more detailed evaluation of the binding energetics of MAO-A inhibitors, we explored the behavior of the predicted and solvated poses in a machine learning model. This model was constructed in KNIME based on data obtained previously for >9000 solvated protein-ligand complexes. The model is still preliminary, but we show that it may be able to predict affinity to within about 1 kcal/mol. This model was used to attempt a prediction of the affinities of the tested ligands with MAO-A.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

NMI (near-infrared dye conjugate MAO A inhibitor) outperformed FDA-approved prostate cancer drugs with a unique mechanism based on bioinformatic analysis of NCI60 screening data

PDF

Molecular docking of sulfonylureas to the SUR1 receptor

PDF

Potential therapeutic effect of monoamine oxidase (MAO) inhibitor on human neuroblastoma

PDF

Integration of KNIME and molecular docking for evaluation of tau fibril inhibitors

PDF

NMI: a near infrared conjugated MAO-A inhibitor as a novel targeted therapy for colorectal and other cancers

PDF

Modeling of water molecules in protein-ligand binding

PDF

Computational model for predicting ionic solubility

PDF

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

PDF

Bioinformatics analysis of the anti-cancer potency of NMI on non-small cell lung cancer and its potential mechanism

PDF

Development of a high-throughput screening assay to study mPD-1/mPD-L1 interactions

PDF

Monoamine oxidase (MAO) knock-out mouse models: Tools for studying the molecular basis of aggression, anxiety, autism and cancers

PDF

Artificial intelligence in medicinal chemistry and drug discovery

PDF

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

PDF

Predicting mortality of sepsis with machine learning model approaches

PDF

Pharmacokinetic modeling: ciprofloxacin in the environment and metformin PBPK model

PDF

Assessment of theranostic agent for brain cancer

PDF

Heat shock proteins and their role in cancer treatment

PDF

Rational design of monoclonal antibody inhibitors targeting human cathepsin B

PDF

Cyclotides: discovery, structure, synthesis and applications

Asset Metadata

Creator Joshi, Sawanee Abhijit (author)

Core Title Inhibition of MAO-A by Dual MAO-A/HDAC inhibitors: in silico approach for ligand binding and affinity prediction

School School of Pharmacy

Degree Master of Science

Degree Program Pharmaceutical Sciences

Degree Conferral Date 2023-08

Publication Date 08/14/2023

Defense Date 09/09/2022

Publisher University of Southern California. Libraries (digital)

Tag dual inhibitors,in-silico,MAO-A,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Haworth, Ian (committee chair), Shih, Jean (committee member), Duncan, Roger (committee member)

Creator Email sajoshi@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113298126

Unique identifier UC113298126

Identifier etd-JoshiSawan-12252.pdf (filename)

Document Type Thesis

Rights Joshi, Sawanee Abhijit

Internet Media Type application/pdf

Type texts

Source 20230816-usctheses-batch-1085 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA

Repository Email cisadmin@lib.usc.edu