QSAR Classification-Based Virtual Screening Followed by Molecular Docking Identification of Potential COX-2 Inhibitors in a Natural Product Library

Abstract

Developments of natural inhibitors to prevent the function of cyclooxygenase-2 (COX-2) protein, responsible for a variety of inflammations and cancers, are a major challenge in the scientific community. In this study, robust QSAR classification models for predicting COX-2 inhibitor were developed, by which the self-organizing feature map neural network and random forest (RF) were adopted to improve the prediction of classification model ability. The F-score-based criterion combined with RF was used for feature selection, and good performance for COX-2 inhibitor prediction in overall accuracy was demonstrated. We used this model as a virtual screening tool for identifying the potential COX-2 inhibitor from a natural product library and found potential hit compounds. This compound further screened by applying molecular docking simulation identified five potential hits such as osthole, kavain, vanillyl acetone, myristicin, and psoralen, having a comparable binding affinity to COX-2 protein. However, in cell experiment, three hit compounds revealed COX-2 inhibitory activity in mRNA and protein level such as osthole, kavain, and psoralen.

1. Introduction

COX Cyclooxygenase, well known as prostaglandin endoperoxide synthase, is responsible for the formation of prostanoids. Cyclooxygenase-2 (COX-2) involves in many diseases as a COX subtype, such as inflammation and many cancers (Limongelli et al., 2010). Nonsteroidal anti-inflammatory drugs (NSAIDs) selective for inhibition of COX-2 are likely without gastrointestinal side effects, but able to cause cardiovascular events, including heart failure, myocardial infarction, and stroke (Micklewright et al., 2003; Hegazy and Ali, 2012). Discovering a drug for inhibition of COX-2 remains promising despite COX-2 being inhibited by natural products such as flavonoids, fish oils, and vitamins, which provide less cardiovascular risk than NSAIDs (Kingston, 2011).

In the past year, to find a better lead compound, ligand-based drug designing (LBDD) (Franke et al., 2007; Aparoy et al., 2010) and structure-based drug designing (SBDD) have been performed as a significant role in drug discovery. LBDD based on pharmacophore and QSAR (quantitative structure/activity relationship) models (Franke et al., 2007; Aparoy et al., 2010) gained by account the knowledge of known inhibitors that bind to the biological target of interest, while SBDD is used for the development of its inhibitor on structural features of the drug target. QSAR models are essential for LBDD that extract quantitative relationships between the biological activities of compounds and their chemical structures, which formulate to an appropriate model to predicting inhibitors such as 5-lipoxygenase and COX-2 (Zheng et al., 2006; Aparoy et al., 2010).

Virtual screening is a computational technique used in drug discovery to identify those structures in small-molecule library that are most likely to impact drug targets. The automatically evaluating very large chemical library on computation based on molecular descriptors with machine-learning algorithms in virtual screening is a powerful tool in drug discovery. QSAR scheme in digital form is dependent on molecular descriptors acquired from chemical information, providing links between molecular structure, physicochemical properties, and quantity. QSAR usually used in the identification of chemical structure is a specific inhibitor of the target protein by mathematical quantification and computerization. Machine-learning techniques have been applied to advancing cheminformatics disciplines for developing QSAR to more precision by which machine learning techniques such as artificial neural network, k-nearest neighbors, support vector machine (SVM), random forest (RF), or decision trees enable discovery of more complex nonlinear relationships in experimental data.

To the best of our knowledge, there is no QSAR model developed for COX-2 inhibitor by machine-learning techniques to virtual screen in large numbers of molecules. Therefore, our study is intended to develop QSAR models by the self-organizing feature map (SOM) neural network and RF to virtual screening of COX-2 inhibitors in a natural product library and to determine this method's accuracy in mRNA and protein level using real time (RT)-polymerase chain reaction (RT-PCR) and Western blotting.

2. Materials and Methods

2.1. Data set

The data set used in this study, derived from other publications (Kauffman and Jurs, 2010), contains 176 compounds that possess a COX-2 inhibition activity span of IC₅₀ value, and these compounds range from 1 nM to 100 μM (Fig. 1 and Table 1). We divided high- and low-activity compound by IC₅₀ = 6.5, referred to as the standard.

FIG. 1.

Scaffold chemical structure.

Table 1.

Structure, Experimental Activity of Analogs

No.	Scaffold chemical	Radical group				IC₅₀ (μm)	Activity
No.	Scaffold chemical	R1	R2	R3	R4	IC₅₀ (μm)	Activity
1	a	-4-SO₂CH₃	-4-F	-CH₃	-H	0.06	High
2	a	-4-SO₂CH₃	-H	-CH₃	-H	0.06	High
3	a	-4-SO₂CH₃	-4-CF₃	-CH₃	-H	0.08	High
4	a	-4-SO₂CH₃	-4-CH₃	-CH₃	-H	0.04	High
5	a	-4-SO₂CH₃	-4-F	-CH₂CH₃	-H	>100	Low
6	a	-4-SO₂CH₃	-3-SO₂CH₃	-H	-H	>100	Low
7	a	-4-SO₂CH₃	-4-SO₂CH₂CH₃	-H	-H	>100	Low
8	a	-4-SO₂CH₃	-4-SO₂C₆H₅	-H	-H	>100	Low
9	a	-4-SO₂CH₃	-3-Cl-4-SO₂CH₃	-H	-H	>100	Low
10	a	-4-F	-4-SO₂NH₂	-H	-H	0.014	High
11	a	-4-SO₂CH₃	-4-SO₂NHCH₃	-H	-H	>100	Low
12	a	-4-SO₂CH₃	-4-SO₂N(CH₃)₂	-H	-H	>100	Low
13	a	-4-SO₂CH₃	-4-F	-CH₂CH₂CO₂H	-H	>100	Low
14	a	-4-SO₂CH₃	-4-F	-CF₃	-H	>100	Low
15	a	-4-SO₂CH₃	-4-F	-CH₃	-SOCF₃	0.06	High
16	a	-4-SO₂CH₃	-4-F	-CH₃	-Br	0.02	High
17	a	-4-SO₂CH₃	-4-F	-CH₃	-Cl	0.05	High
18	a	-4-SO₂CH₃	-4-F	-CH₃	-CH₂N(CH₃)₂	>100	Low
19	a	-4-SO₂CH₃	-4-F	-CH₃	-CH₂OC₆H₄-4-Cl	0.03	High
20	a	-4-SO₂CH₃	-4-F	-CH₃	-CH₂OC₆H₄-3-Cl	0.08	High
21	b	-4-SO₂CH₃	-4-NHCH₃	-CF₃	^*	1.47	Low
22	b	-4-SO₂CH₃	-4-N(CH₃)₂	-CF₃	^*	0.7	Low
23	b	-4-SO₂CH₃	-4-SO₂CH₃	-CF₃	^*	5.7	Low
24	b	-4-SO₂NH₂	-4-Cl	-CF₃	^*	0.01	High
25	b	-4-SO₂NH₂	-4-F	-CF₃	^*	0.01	High
26	b	-4-SO₂NH₂	-H	-CF₃	^*	0.04	High
27	b	-4-SO₂NH₂	-4-CH₃	-CF₃	^*	0.04	High
28	b	-4-SO₂CH₃	-3-Cl	-CF₃	^*	0.06	High
29	b	-4-SO₂CH₃	-3-Br	-CF₃	^*	0.08	High
30	b	-4-SO₂CH₃	-3-CH₃	-CF₃	^*	0.06	High
31	b	-4-SO₂CH₃	-3-OCH₃	-CF₃	^*	0.35	Low
32	b	-4-SO₂CH₃	-3-SCH₃	-CF₃	^*	0.35	Low
33	b	-4-SO₂CH₃	-3-CH₂OCH₃	-CF₃	^*	68.1	Low
34	b	-4-SO₂CH₃	-3-N(CH₃)₂	-CF₃	^*	3.2	Low
35	b	-4-SO₂CH₃	-3-NHCH₃	-CF₃	^*	0.92	Low
36	b	-4-SO₂CH₃	-3-NH₂	-CF₃	^*	5.89	Low
37	b	-4-SO₂NH₂	-3-F	-CF₃	^*	0.03	High
38	b	-4-SO₂NH₂	-3-CH₃	-CF₃	^*	0.03	High
39	b	-4-SO₂CH₃	-2-Cl	-CF₃	^*	0.9	Low
40	b	-4-SO₂CH₃	-2-F	-CF₃	^*	0.4	Low
41	b	-4-SO₂CH₃	-2-CH₃	-CF₃	^*	0.8	Low
42	b	-4-SO₂CH₃	-2-OCH₃	-CF₃	^*	>100	Low
43	b	-4-SO₂CH₃	-4-SCH₃-3-Cl	-CF₃	^*	0.04	High
44	b	-4-SO₂CH₃	-4-CH₃-3-Cl	-CF₃	^*	0.03	High
45	b	-4-SO₂CH₃	-3-CH₃-4-Cl	-CF₃	^*	0.09	High
46	b	-4-SO₂CH₃	-3-N(CH₃)₂-4-Cl	-CF₃	^*	1.04	Low
47	b	-4-SO₂CH₃	-3,4- CH₃	-CF₃	^*	0.33	Low
48	b	-4-SO₂CH₃	-3-CH₃-5-Cl	-CF₃	^*	0.08	High
49	b	-4-SO₂CH₃	-3-OCH₃-5-F	-CF₃	^*	0.96	Low
50	b	-4-SO₂CH₃	-3-CF₃-5-F	-CF₃	^*	>100	Low
51	b	-4-SO₂CH₃	-2-CH₃-5-F	-CF₃	^*	>100	Low
52	b	-4-SO₂CH₃	-2-CH₃-6-Cl	-CF₃	^*	>100	Low
53	b	-4-SO₂NH₂	-4-OCH₃-3-F	-CF₃	^*	0.03	High
54	b	-4-SO₂NH₂	-4-OCH₃-3-Cl	-CF₃	^*	0.02	High
55	b	-4-SO₂NH₂	-4-OCH₃-3-Br	-CF₃	^*	0.03	High
56	b	-4-SO₂NH₂	-4-SCH₃-3-Cl	-CF₃	^*	0.01	High
57	b	-4-SO₂NH₂	-4-CH₃-3-Cl	-CF₃	^*	0.003	High
58	b	-4-SO₂CH₃	-4-OCH₃-3,5-Cl	-CF₃	^*	0.14	High
59	b	-4-SO₂NH₂	-3,4-F	-CF₃	^*	0.03	High
60	b	-4-SO₂NH₂	-3-CH₃-5-Cl	-CF₃	^*	0.04	High
61	b	-4-SO₂NH₂	-3-CH₃-5-F	-CF₃	^*	0.03	High
62	b	-4-SO₂NH₂	-3-OCH₃-5-Cl	-CF₃	^*	0.46	Low
63	b	-4-SO₂CH₃	-4-OCH₃-3,5-Br	-CF₃	^*	0.09	High
64	b	-4-SO₂CH₃	-4-OCH₃-3,5-CH₃	-CF₃	^*	0.72	Low
65	b	-4-SO₂CH₃	-4-OCH₃-2,5-CH₃	-CF₃	^*	12.2	Low
66	b	-4-SO₂CH₃	-4-N(CH₃)₂-3,5-F	-CF₃	^*	0.03	High
67	b	-4-SO₂CH₃	-4-Cl	-H	^*	>100	Low
68	b	-4-SO₂CH₃	-4-Cl	-CHF₂	^*	0.61	Low
69	b	-4-SO₂CH₃	-4-Cl	-CO₂H	^*	>100	Low
70	b	-4-SO₂CH₃	-4-Cl	-CO₂N(CH₂CH₃)₂	^*	>100	Low
71	b	-4-SO₂CH₃	-4-Cl	-CH₂NH₂	^*	>100	Low
72	b	-4-SO₂CH₃	-4-Cl	-CH₂OCH₃	^*	3.72	Low
73	b	-4-SO₂CH₃	-4-Cl	-CH₂SO₂CH₃	^*	>100	Low
74	b	-4-SO₂CH₃	-4-Cl	-CH₂OH	^*	8.35	Low
75	c	-4-SO₂CH₃	-4-F	-H	^*	0.026	High
76	c	-4-SO₂CH₃	-4-OCH₃	-H	^*	0.005	High
77	c	-4-SO₂CH₃	-4-Cl	-H	^*	0.003	High
78	c	-4-SO₂CH₃	-4-CH₃	-H	^*	0.003	High
79	c	-4-SO₂CH₃	-H	-H	^*	2.25	Low
80	c	-4-SO₂CH₃	-4-CF₃	-H	^*	0.865	Low
81	c	-4-SO₂CH₃	-2,4-Cl	-H	^*	0.053	High
82	c	-4-SO₂CH₃	-4-CH₂OH	-H	^*	3.2	Low
83	c	-4-SO₂CH₃	-4-CH₂OCH₃	-H	^*	6.6	Low
84	c	-4-SO₂CH₃	-4-F	-CH₃	^*	0.015	High
85	c	-4-SO₂CH₃	-4-Cl	-CH₃	^*	0.007	High
86	c	-4-SO₂CH₃	-4-F	-C₂H₅	^*	65	Low
87	c	-4-SO₂CH₃	-4-F	-CH₂F	^*	0.051	High
88	c	-4-SO₂CH₃	-3,4-F	-H	^*	0.051	High
89	c	-4-SO₂CH₃	-4-F-3-Cl	-H	^*	0.03	High
90	c	-4-SO₂CH₃	-3-CF₃-4-F	-H	^*	>100	Low
91	c	-4-SO₂CH₃	-3,4,5-F	-H	^*	>100	Low
92	c	-4-SO₂CH₃	-3,4-Cl	-H	^*	0.01	High
93	c	-4-SO₂CH₃	-4-OCH₃-3,4-Cl	-H	^*	0.017	High
94	c	-4-SO₂CH₃	-4-N(CH₃)₂-3-Cl	-H	^*	0.005	High
95	c	-4-SO₂NH₂	-4-F	-H	^*	0.007	High
96	c	-4-SO₂NH₂	-3,4-F	-H	^*	0.018	High
97	c	-4-SO₂NH₂	-4-F-3-Cl	-H	^*	0.01	High
98	c	-4-SO₂NH₂	-3,4,5-F	-H	^*	2.9	Low
99	c	-4-SO₂NH₂	-4-Cl	-H	^*	0.003	High
100	c	-4-SO₂NH₂	-3,4-Cl	-H	^*	0.002	High
101	c	-4-SO₂NH₂	-4-OCH₃	-H	^*	0.002	High
102	c	-4-SO₂NH₂	-4-OCH₃-3-F	-H	^*	0.016	High
103	c	-4-SO₂NH₂	-4-OCH₃-3-Cl	-H	^*	0.009	High
104	c	-4-SO₂NH₂	-4-OCH₃-3,5-Cl	-H	^*	0.009	High
105	c	-4-SO₂NH₂	-4-N(CH₃)₂-3-Cl	-H	^*	0.002	High
106	d	-4-SO₂CH₃	-4-OCH₃-3-F	-H	^*	>100	Low
107	d	-4-SO₂NH₂	-4-F-3-Cl	-H	^*	0.017	High
108	d	-4-SO₂NH₂	-4-Cl	-H	^*	0.006	High
109	d	-4-SO₂NH₂	-4-OCH₃-3-Cl	-H	^*	0.019	High
110	d	-4-SO₂CH₃	-4-F	-3,4-F	^*	0.014	High
111	d	-4-SO₂CH₃	-4-F-3-Cl	-3,4-F	^*	0.01	High
112	d	-4-SO₂CH₃	-4-F-3-CH₃	-3,4-F	^*	0.005	High
113	d	-4-SO₂CH₃	-4-OCH₃-3-Cl	-3,4-F	^*	0.019	High
114	d	-4-SO₂CH₃	-4-OCH₃-3,5-Cl	-3,4-F	^*	>100	Low
115	d	-4-SO₂CH₃	-3,4-OCH₃	-3,4-F	^*	0.34	Low
116	d	-4-SO₂CH₃	-4-CH₃	-3,4-F	^*	0.007	High
117	d	-4-SO₂CH₃	-4-Cl-3-CH₃	-3,4-F	^*	0.006	High
118	d	-4-SO₂CH₃	-4-N(CH₃)₂-3-Cl	-3,4-F	^*	0.008	High
119	d	-4-SO₂NH₂	-4-F	-3,4-F	^*	0.004	High
120	d	-4-SO₂NH₂	-4-F-3-Cl	-3,4-F	^*	0.002	High
121	d	-4-SO₂NH₂	-4-F-3-CH₃	-3,4-F	^*	0.002	High
122	d	-4-SO₂NH₂	-4-OCH₃-3-F	-3,4-F	^*	0.013	High
123	d	-4-SO₂NH₂	-4-OCH₃-3-Cl	-3,4-F	^*	0.013	High
124	d	-4-SO₂NH₂	-4-OCH₃-3-CH₃	-3,4-F	^*	0.005	High
125	d	-4-SO₂NH₂	-4-CH₃	-3,4-F	^*	0.004	High
126	d	-4-SO₂NH₂	-4-CH₃-3-Cl	-3,4-F	^*	0.003	High
127	d	-4-SO₂CH₃	-3,4-CH₃	-3,4-F	^*	0.005	High
128	d	-4-SO₂CH₃	-4-F	-2,3,4,5-F	^*	>100	Low
129	e	-4-SO₂NH₂	-2-Cl	-CF₃	-H	0.01	High
130	e	-4-SO₂NH₂	-4-CH₂CH₃	-CF₃	-H	0.86	Low
131	e	-4-SO₂NH₂	-4-CF₃	-CF₃	-H	8.23	Low
132	e	-4-SO₂NH₂	-4-OH	-CF₃	-H	>100	Low
133	e	-4-SO₂NH₂	-4-OCH₃	-CF₃	-H	0.008	High
134	e	-4-SO₂NH₂	-4-OCH₂CH₃	-CF₃	-H	0.64	Low
135	e	-4-SO₂NH₂	-4-SCH₃	-CF₃	-H	0.009	High
136	e	-4-SO₂NH₂	-2-N(CH₃)₂	-CF₃	-H	14.3	Low
137	e	-4-SO₂NH₂	-4-N(CH₃)₂	-CF₃	-H	0.005	High
138	e	-4-SO₂NH₂	-4-CH₂OH	-CF₃	-H	93.3	Low
139	e	-4-SO₂NH₂	-4-CO₂H	-CF₃	-H	11.2	Low
140	e	-4-SO₂NH₂	-2,5-Cl	-CF₃	-H	>100	Low
141	e	-4-SO₂NH₂	-4-Cl	-CHF₂	-H	0.01	High
142	e	-4-SO₂NH₂	-4-SO₂CH₃	-CHF₂	-H	>100	Low
143	e	-4-SO₂NH₂	-4-CONH₂	-CHF₂	-H	>100	Low
144	e	-4-SO₂NH₂	-4-CO₂H	-CHF₂	-H	46.8	Low
145	e	-4-SO₂NH₂	-4-OCH₃	-CHF₂	-H	0.015	High
146	e	-4-SO₂NH₂	-2,5-CH₃	-CHF₂	-H	>100	Low
147	e	-4-SO₂NH₂	-H	-CH₃	-H	62.8	Low
148	e	-4-SO₂NH₂	-4-F	-CN	-H	0.34	Low
149	e	-4-SO₂NH₂	-4-Cl	-CH₂OH	-H	0.83	Low
150	e	-4-SO₂NH₂	-H	-OCH₃	-H	>100	Low
151	e	-4-SO₂NH₂	-4-Cl	-CN	-Cl	0.01	High
152	e	-4-SO₂NH₂	-4-F	-H	-H	>100	Low
153	e	-4-SO₂NH₂	-4-Cl	-H	-SO₂CH₃	19.8	Low
154	e	-4-SO₂NH₂	-H	-H	-NH₂	29.8	Low
155	e	-4-SO₂NH₂	-H	-CF₃	-F	0.002	High
156	e	-4-SO₂NH₂	-H	-CF₃	-OH	3.58	Low
157	e	-4-F	-4-SO₂NH₂	-CF₃	-H	0.01	High
158	e	-4-SO₂N(CH₃)₂	-4-Cl	-CF₃	-H	>100	Low
159	e	-4-NOCF₃	-4-Cl	-CF₃	-H	>100	Low
160	e	-H	-H	-CF₃	-H	>100	Low
161	e	-4-Cl	-4-Cl	-CF₃	-H	4.79	Low
162	e	-4-OCH₃	-4-OCH₃	-CF₃	-H	0.75	Low
163	e	-4-OCH₃	-4-Cl	-CF₃	-H	74.9	Low
164	f	-H	-4-SO₂CH₃	-C₆H₄-2-Cl	^*	0.01	High
165	f	-2-F	-4-SO₂CH₃	-C₆H₄-2-Cl	^*	0.01	High
166	f	-2,5-F	-4-SO₂CH₃	-C₆H₄-2-Cl	^*	>100	Low
167	f	-4-F	-4-SO₂NH₂	-C₆H₄-2-Cl	^*	0.013	High
168	f	-4-Cl	-4-SO₂NH₂	-CH₃	^*	0.005	High
169	f	-4-OCH₃-3-F	-4-SO₂NH₂	-C₆H₄-2-Cl	^*	2.48	Low
170	g	-4-SO₂CH₃	-4-F	-H	-H	>100	Low
171	g	-4-SO₂CH₃	-4-F	-CF₃	-H	>100	Low
172	g	-4-SO₂CH₃	-4-F	-H	-CH₂CH = CH₂	>100	Low
173	h	-4-SO₂CH₃	-4-F	^*	^*	0.02	High
174	h	-4-SO₂NH₂	-4-F	^*	^*	0.007	High
175	h	-4-SO₂NH₂	-3-F-4-OCH₃	^*	^*	0.005	High
176	h	-4-SO₂NH₂	-3-Cl-4-OCH₃	^*	^*	0.02	High

2.2. Molecular modeling and descriptor calculation

All the structures in the database were subsequently converted to SMILES (Simplified Molecular Input Line Entry System) using ChemDraw V.10.0 (Mills, 2006), and the 176 molecular descriptors classified as 0D, 1D, and 2D families were calculated by Mold 2 (Dong et al., 2015).

2.3. Feature selection method

A selection criterion is required to remove irrelevant descriptors for the selection of suitable descriptors from a very large number of raw descriptors that contain little information or correlated with other descriptors without causing much loss of information (Jović et al., 2015). In this study, the F-score-based criterion was adopted; if the numbers of positive and negative instances are n₊, and n₋, respectively, then the F-score of the ith feature is defined as follows:

\documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*} { F_i } = { \frac { { { \left( { \bar x_i^ { ( + ) } - { { \bar x } _i } } \right) } ^2 } + { { \left( { \bar x_i^ { ( - ) } - { { \bar x } _i } } \right) } ^2 } } { \frac { 1 } { { { n_ + } - 1 } } \mathop \sum \nolimits_ { k = 1 } ^ { { n_ + } } { { \left( { x_ { k , \;i } ^ { \left( + \right) } - \bar x_i^ { ( + ) } } \right) } ^2 } + \frac { 1 } { { { n_ - } - 1 } } \mathop \sum \nolimits_ { k = 1 } ^ { { n_ - } } { { \left( { x_ { k , \;i } ^ { \left( - \right) } - \bar x_i^ { ( - ) } } \right) } ^2 } } } \tag { 1 } \end{align*} \end{document}

where \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $${ \bar x_i}$$ \end{document} , \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$\bar x_i^{ ( + ) }$$ \end{document} , and \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$\bar x_i^{ ( - ) }$$ \end{document} denote the average of the ith feature of the whole, positive, and negative data sets, respectively; \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $${ \rm{ \;}}x_{k , \;i}^{ \left( + \right) }$$ \end{document} denotes the ith feature of the kth positive samples, and \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$x_{k , \;i}^{ \left( - \right) }$$ \end{document} denotes the ith feature of the kth negative samples. The numerator in (1) indicates the discrimination between the positive and negative sets and the denominator indicates the one within each of the two sets.

2.4. QSAR modeling

QSAR classification models have correlativity between independent variables (molecular descriptors) and a response variable that represents the class of the corresponding sample (COX-2 inhibition activity). Machine-learning techniques are generally conducted for grouping observations or instance into classes in QSAR modeling where the relationship between structure and activity is often complex and nonlinear (Lavecchia, 2015).

In this study, both the SOM neural network and RF were used. SOM is a kind of self-organization and self-learning network, which learns and classifies the input vectors in the input space as well as learns the distribution characteristics and topological structure of the input vectors (Vesanto and Alhoniemi, 2000). SOM neural network is used to distribute the input training set samples to neurons for realizing the data clustering that could randomly select some data from each class as the training set. RF algorithm combines the idea of bootstrap, aggregating with the random subspace methods that are a classifier containing multiple decision trees randomly generated. RF algorithm is similar to a bagging algorithm in resampling based on a bootstrap method to produce multiple training sets, and a decision tree trained with a training set generated by bootstrap (Breiman, 2001).

2.5. Model performance evaluation

In this study, for assessment of accuracy and selection of optimal model, performance of classification model was evaluated by receiver operating curves (ROCs). Several metrics derived from the confusion matrix were used, including sensitivity, SE = TP/(TP + FN), specificity, SP = FP/(FP + TN), and overall prediction accuracy, CA = (TP + TN)/(TP + TN + FP + FN) (TP denotes true positives; FN denotes false negatives; FP denotes false positives; and TN denotes true negatives, respectively).

CA is the total percentage of both high and low compounds that is correctly predicted; sensitivity is the proportion of actual active compound that is correctly identified; specificity is the proportion of actual inactive compound that is correctly identified; ROC have measured the quality of the classification models.

2.6. Virtual screening

Virtual screening is to find potential leads with different scaffolds from a chemical library by QSAR models. The chemical library consisted of 502 chemicals derived from natural plants and microorganisms.

2.7. Molecular docking

Molecular docking has been used to predict the best binding ligand molecule on the active site of receptor targets as well as virtual screening as a computational tool. Three-dimensional (3D) structural information obtained from the Protein Data Bank, Human COX-2 stable crystal structure (PDB ID: 5ikq), for the molecular docking applied by which MOE (Molecular Operating Environment, v2014.0901; Chemical Computing Group, Inc.) software determined the overall lowest potential energy configuration by protonating 3D module in the different states of terminal amides, hydroxyls, thiols, histidines, and titratable groups.

2.8. Chemicals

Extracted chemicals from natural products, osthole, myristicin, and vanillyl acetone, were purchased from Yuanye Biology (Beijing, China), and psoralen and kavain were purchased from Beijing Zhongke Yiyou Chemical Technology Research Institute (Beijing, China).

2.9. Cell line and cell culture

HepG2, human liver cancer line, was obtained from the American Type Culture Collection (ATCC, Manassas, VA) and cultured in Dulbecco's modified Eagle's medium (DMEM; Thermo Fisher Scientific, Waltham, MA), supplemented with 10% fetal bovine serum (Thermo Fisher Scientific). The cells were incubated at 37°C in a humidified atmosphere with 5% carbon dioxide.

2.10. MTT 3-(4,5-dimethyl thiazole-2-yl)-2,5-diphenyl tetrazolium bromide assay

HepG2 cells were treated with osthole (0, 40, 80, 120, 160, 200 μM), psoralen (0, 10, 50, 100, 150, 200 μM), kavain (0, 100, 200, 300, 400, 500 μM), myristicin (0, 5, 6, 7, 8, 10 mM), and vanillyl acetone (0, 1.1, 1.2, 1.4, 1.6, 2.0 mM) for 48 hours, and the absorbance was measured at 570 nm after adding the MTT 3-(4,5-dimethyl thiazole-2-yl)-2,5-diphenyl tetrazolium bromide assay (Sigma-Aldrich, St. Louis, MO) for 4 hours at 37°C.

2.11. Real-time polymerase chain reaction

Total RNA was extracted from cells using the TRIzol reagent (Invitrogen, Thermo Fisher Scientific), and then, cDNA was synthesized using the PrimeScript™ RT Reagent Kit, and gDNA Eraser (Perfect Real TIME, Takara, Japan) for what real-time PCR was carried out using SuperReal PreMix Plus (SYBR Green; Tiangen, China). The PCR primers were as following: COX-2 (forward: TCACAGGCTTCCATTGACCAG; reverse: CCGAGGCTTTTCTACCAGA), GAPDH (forward: AATCCCATCACCATCTTCC; reverse: CATCACGCCACAGTTTCC), and the relative expression of mRNA was calculated by the 2^{(−delta CT)} method.

2.12. Western blotting

COX-2 and β-actin antibody were purchased from Cell Signaling Technology (Danvers, MA) and Abcam (Cambridge, United Kingdom), and the relative density was calculated as the following formula: COX-2/β-actin band density.

2.13. Statistical analysis

The results were provided as mean ± SD, a difference between groups assessed through one-way ANOVA by SPSS 20.0, and p < 0.05 assumed as statistically significant.

3. Results and Discussion

3.1. Classification models for cyclooxygenase-2 inhibitor

The RF algorithm combines the idea of bootstrap, which aggregating with the random subspace methods that are a classifier contains multiple decision trees randomly generated. As the test data input to the RF classifier, each decision tree makes the decision for every training set, and the final classification acquired by majority vote. The RF algorithm is described in the flow chart in Figure 2.

FIG. 2.

Flow chart of the RF algorithm. RF, random forest.

To reduce the redundant variables, data preprocessing is implemented as follows: (1) delete the molecular descriptors that contain zero values higher than 85%; (2) remove the molecular descriptors where invariance is lower than 0.05; and (3) remove the pair of molecular descriptors where the correlation coefficient is higher than 0.95. The number of molecular descriptor variables dropped to 180 for further investigation after preprocessing, and the F-score algorithm reduced the dimension of the 180 molecular descriptors to obtain an optimal molecular descriptor subset. The flow chart of obtaining the optimal subset of molecular descriptors is shown in Figure 3. The number of molecular descriptors chosen in the feature selection, N, was decided by looping the entire computational procedure for modeling. The number of molecular descriptors reduced to N = 110 for the feature selection, and the optimal subset of molecular descriptors in size N is obtained.

FIG. 3.

Flow chart of obtaining the optimal molecular descriptor subset.

The training and test set divided by which the SOM neural network clustered the 176 molecular descriptors into 16 neurons on the map by grids of 4 × 4 size (Fig. 4). To ensure the training set uniformly distributed in the data space to build a proper model, the selected random number was labeled in black from each cluster as the training set and the rest labeled in red as the test set.

FIG. 4.

Split training and test sets by the SOM neural network. SOM, self-organizing feature map.

To more likely predict the COX-2 inhibitor, feature selection (FS) combined with RF showed good performance for COX-2 inhibitor prediction in SE, SP, and Q, respectively, in FS-RF versus RF (Table 2). In addition, area under the curve (AUC) \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$ { \rm { \;AUC } } = \mathop \smallint \int \limits_0^1 { \frac { { \rm { TP } } } { \rm { P } } } { \rm { d } } { \frac { { \rm { FP } } } { \rm { N } } } $$ \end{document} (Table 2) value and ROC were used for evaluating the prediction performance (Fig. 5).

FIG. 5.

The ROC of FS-RF and RF. FS-RF, feature selection-random forest; ROC, receiver operating curve.

Table 2.

The Prediction Performance of COX-2 Inhibitors Using Random Forest (RF) and FS-Random Forest (Feature Selection-Random Forest) Methods

Model	SE, %	SP, %	Q, %	AUC
RF	73.81	76.67	75	0.7374
FS-RF	88.24	87.5	87.88	0.8735

AUC, area under the curve; COX-2, cyclooxygenase-2; FS-RF, feature selection-random forest.

3.2. Virtual screening

Recently, to prediction the lead compound in drug discovery, many different strategies, including ligand-based approaches (SAR, QSAR, pharmacophore mapping), have been performed, and which further supervised (SVM, neural network, decision tree, RF, kappa nearest neighbor, naive Bayesian, binary QSAR) and unsupervised (self-organizing map) learning methods. In our study, a combination of RF and SOM neural network is applied for developing machine learning to improve the virtual screen. We screened from the natural product library 502 chemicals acquired from natural plants and other sources (Table 3). Each of the 502 chemical structures from natural products shifted to molecular descriptors by mold2, and chemicals were identified as COX-2 inhibitor by the FS-RF model to further molecular docking.

Table 3.

Chemical Library for Virtual Screening from Nature Products

No.	Name	No.	Name	No.	Name
1	Acivicin	2	Actinomycin D	3	Anisomycin
4	Antibiotic A-23187	5	Aristolochic acid A	6	Artesunate
7	Australine HCl	8	Baicalein	9	Betulinic acid
10	Bilobalide	11	Brefeldin A	12	Bromocriptine mesylate
13	C2 Phytoceramide	14	C6 Ceramide	15	Caffeic acid
16	Camptothecin	17	Cantharidin	18	CAPE
19	Capsaicin	20	Castanospermine	21	Cerulenin
22	Cevadine	23	Chaetomellic acid A	24	Chelerythrine Cl
25	Chromomycin A3	26	Citrinin	27	Colchicine
28	Coumermycin A1	29	Curcumin	30	Cycloheximide
31	Cyclopamine	32	Cyclopiazonic acid	33	Cycloserine, L-
34	Cyclosporin A	35	Cytochalasin B	36	Cytochalasin D
37	Cytochalasin E	38	Daidzein	39	Daunorubicin HCl
40	Decoyinine	41	Deguelin	42	Deoxyphorbol 13-acetate, 12-
43	Deoxyphorbol 13-phenylacetate 20-acetate, 12-	44	Dihydroergocristine mesylate	45	Domoic acid
46	Doxorubicin HCl	47	E6 Berbamine	48	E-64
49	E-64-C	50	E-64-D	51	Ebelactone B
52	Ellipticine	53	Embelin	54	Epibatidine, (±)-
55	Epigallocatechin gallate	56	Etoposide	57	Forskolin
58	Fumagallin	59	Fumonisin B2	60	Galantamine HBr
61	Gambogic acid	62	Genistein	63	Geranylgeranoic acid
64	Gingerol	65	Ginkgolide B	66	Gliotoxin
67	Gossypol	68	Grayanotoxin III	69	Himbacine
70	Huperzine A, (−)-	71	Hydroxycamptothecin, 10-	72	Hypericin
73	Indirubin	74	Ingenol 3,20-dibenzoate	75	Isotetrandrine
76	Jervine	77	Kainic acid	78	Kavain (±)
79	Kenpaullone	80	Lapachone, b-	81	Lincomycin
82	Lycorine HCl	83	Mevastatin	84	Mezerein
85	Mimosine, L-	86	Mithramycin A	87	Monensin Na
88	Mycophenolic acid	89	Myriocin	90	Neomycin sulfate
91	Nigericin Na	92	Oligomycin A	93	Ouabain (−)-
94	Parthenolide	95	Perillic acid	96	Phloretin
97	Phorbol 12,13-dibutyrate	98	Guaiol	99	Phorbol 12-myristate 13-acetate
100	Phorbol 12-myristate 13-acetate, 4-a-	101	Phytosphingosine	102	Piceatannol
103	Prostaglandin A1	104	Prostaglandin B1	105	Prostaglandin E1
106	Prostaglandin E2	107	Prostaglandin F2a	108	Prostaglandin J2
109	Quisqualic acid	110	Radicicol	111	Rapamycin
112	Rauwolscine	113	Resveratrol	114	Retinoic acid (all trans)
115	Retinoic acid, 13-cis-	116	Retinoic acid, 9-cis-	117	Rifampicin
118	Rosmarinic acid	119	Rotenone	120	Rottlerin
121	Ryanodine	122	Shikonin	123	Spectinomycin sulfate
124	Swainsonine	125	Tanshinone IIA	126	Taxol
127	Tetrandrine	128	Thapsigargin	129	Tomatidine
130	Troleandomycin	131	Tunicamycin B	132	Ursolic acid
133	Valinomycin	134	Aconitine	135	Veratridine
136	Vinblastine sulfate	137	Vincristine sulfate	138	Vinpocetine
139	Wedelolactone	140	Wortmannin	141	Apigenin
142	Arecoline HBr	143	Atropine sulfate	144	Berbamine 2HCl
145	Bicuculline, (+)-	146	Bufalin	147	Brucine n-oxide
148	Butein	149	Catalpol	150	Chrysine
151	Desoxypeganine HCl	152	Veratramine	153	Emodin
154	Gramine	155	Harmaline HCl	156	Harmine HCl
157	Hyoscyamine	158	Ivermectin	159	Luteolin
160	Melatonin	161	Morin	162	Myricetin
163	Naringenin	164	Nicotine, (−)-	165	Nonactin
166	Penicillamine, L-	167	Pilocarpine HCl	168	Quassin
169	Picrotoxinin	170	Quercetin	171	Quinidine HCl
172	Quinine HCl	173	Robinetine	174	Menadione
175	Strychnine HCl	176	Tryptanthrin	177	Yohimbine HCl
178	Eburnamonine, (−)-	179	Lysergol	180	Monocrotaline
181	Oxytetracycline, a -apo-	182	Pseudopelletierine HCl	183	Salsolinol HBr
184	Sitosterol, b-	185	Sterigmatocystin	186	Trimethylpsoralen, 4,5′,8-
187	Cinobufagin	188	Emetine 2HCl	189	Kaempferol
190	Kanamycin	191	Kahweol	192	Taxifolin (+)
193	Theobromine	194	Baccatin III	195	Carminic acid
196	Cotinine, (−)-	197	Austricin	198	Condelphine
199	Delcorine	200	Deltaline	201	Diacetyl korseveriline
202	Dubinidine	203	Eudesmine	204	Feroline
205	Fillalbin	206	Graveoline	207	Heliotrine
208	Hernandezine	209	Heteratisine	210	Imperialine
211	Karakoline	212	Lapidine	213	Lapiferine
214	Nitrarine 2HCl	215	Norfluorocurarine	216	Peganole
217	Pinocembrin	218	Protopine HCl	219	Remerine HCl
220	Sevedindione	221	Skimmianine	222	Songorine
223	Trichodesmine	224	Tschimganidin	225	Tschimganine
226	Ungerine nitrate	227	Genistein	228	Amentoflavone
229	Apigenin-7-O-glucoside	230	Bavachinin A	231	Chrysoeriol
232	Convolvamine HCl	233	Daidzein	234	Datiscetin
235	Deacetylcolchicine, N-formyl-	236	Didymin	237	Eriocitrin
238	Eriodictyol	239	Dihydroergotamine methanesulfonate	240	Homobutein
241	Homoeriodictyol (−)	242	Homoorientin	243	Hydroxyflavone, 7-
244	Isorhamnetine-3-O-glucoside	245	Isorhoifoline	246	Isosakuranetin
247	Isovitexin	248	Dihydromethysticin	249	Kaempferol-3-O-glucoside
250	Kaempferol-7-neohesperidoside	251	Luteolin-3′,7-di-o-glucoside	252	Flavokawain B
253	Marein	254	Maritimein	255	Geldanamycin
256	Myricitrin	257	Jasmonic acid	258	Narirutin
259	Picrotin	260	Plumbagin	261	Ketopinic acid
262	Scopolamine N-butylbromide	263	Rhamnetin	264	Rhoifolin
265	Sanguinarine	266	Saponarin	267	Manool
268	Scoulerine	269	Sinensetin	270	Sulfuretin
271	Syrosingopine	272	Tamarixetin	273	Tetrahydroalstonine
274	Methyllycaconitine citrate	275	Vitexin-2”-O-rhamnoside	276	Laudanosoline HBr
277	Bromolaudanosine, (±)-6′-	278	Andrographolide	279	Ajmaline
280	Chelidonine, (+)-	281	Dihydroxyflavone, 6,7-	282	Fisetin
283	Harmalol HCl	284	Harmol HCl	285	Isorhamnetine-3-O-rutinoside
286	Isoscopoletine	287	Methoxy flavone, 5-	288	Syringetine-3-O-glucoside
289	Pratol	290	Conessine	291	Sarsasapogenin
291	Strophanthidin acetate	293	Hordenine sulfate	294	Ferutinin
294	Piperine	296	Quercetin	297	Scopolamine N-oxide HBr, (−)-
297	Shikimic acid	299	Stachydrine HCl	300	Ochratoxin A
301	Patulin	302	Zearalenone	303	Pellitorine
304	Abscisic acid, (±)-	305	Rifamycin SV-NA	306	Aloe-emodine
307	Antimycin A1	308	Asarinin, (−)-	309	Aucubin
310	Deoxyshikonin	311	Boldine	312	Caryophyllene oxide
313	Catechin hydrate, (+)-	314	Cinchonidine, (−)-	315	Cinchonine, (+)-
316	Cotininecarboxylic acid, trans-4-	317	Demissidine	318	Dipterocarpol
319	Distamycin A	320	Friedelin	321	Indole-3-butyric acid
322	Gibberellic acid, (+)-	323	Gitoxigenin	324	Harmane HCl
325	Hydroxytropinone, 6-	326	Isocorydine HCl	327	Isoreserpine, (−)-
328	Leucomisine	329	Methylergonovine	330	Corydaline
331	Muscarine Cl, (+)-	332	Nalidixic acid	333	Narasin
334	Noreleagnine	335	Norharmane	336	Palmatine Cl
337	Peruvoside	338	Physostigmine	339	Uvaol
340	Podocarpic acid	341	Retrorsine	342	Rhapontin
343	Sclareolide, (3aR)-(+)-	344	Streptonigrin	345	Tetrahydropapaverine
346	Ingenol	347	Vineomycin A1	348	Visnagin
349	Wogonin	350	Zearalanol, b-	351	Chaconine, a-
352	Echinomycin	353	Ellagic acid	354	Epicatechin, (−)-
355	Puromycin	356	Glycyrrhetinic acid, 18-b-	357	Griseofulvin, (+)-
358	Isoquercitrin	359	Kinetin	360	Lasalocid A Na
361	Vanillyl acetone	362	Sclareol	363	Trigonelline HCl
364	Tubercidin	365	Usnic acid, (+)-	366	Vitexin
367	Acacetin	368	Capreomycin sulfate	369	Carnitine Cl, (±)-
370	Cephradine	371	Phorbol	372	Homatropine HBr
373	Hydrastine, D-b-	374	Khellin	375	Lobeline HCl
376	Osthole	377	Tetrahydrolipstatin	378	Neohesperidin
379	Noscapine, (±)-	380	Oleanolic acid	381	Papaverine HCl
382	Phlorizin	383	Protoveratrine B	384	Reserpine
385	Salinomycin	386	Xanthotoxin	387	Scopoletin
388	Solanidine	389	Solanine, a-	390	Solasodine
391	Tropine	392	d-Tubocurarine-chloride	393	Myristicin
394	Vincamine	395	Anabasine HCl	396	Andrographolide, dehydro-
397	Dicoumarol	398	Artemisinin	399	Asiatic acid
400	Auraptene	401	Vulpinic acid	402	Berberine HCl
403	Bergenin	404	Biochanin A	405	Bulleyaconitine A
406	Cafestol	407	Cafestol acetate	408	Carnosic acid
409	Catharanthine base	410	Cepharanthine	411	Cryptotanshinone
412	Dehydrokawain, 5,6-	413	Demethylepipodophyllotoxin, 4′-	414	Mitomycin C
415	Methysticin	416	Thymoquinone	417	Dihydrotanshinone
418	Azomycin	419	Diosmetin	420	Diosmin
421	Ecdysone	422	Ecdysone, b-	423	Euphorbia steroid
424	Flavokawain A	425	Lupinine	426	Formononetin
427	Ginkgolide A	428	Harringtonine	429	Hesperetin
430	Hesperidin	431	Honokiol	432	Hypocrellin A
433	Hypocrellin B	434	Lagochiline	435	Lappaconitine
436	Limonin	437	Madecassic acid	438	Magnolol
439	Matrine	440	Minocycline	441	Naringin
442	Nomilin	443	Oxo-cafestol, 16-	444	Oxokahweol, 16-
445	Panaxadiol	446	Panaxadiol	447	Peganine
448	Pimaricin	449	Podophyllotoxin	450	Rubescensin A
451	Rutaecarpine	452	Rutin	453	Salsolidine
454	Salsoline	455	Santonin	456	Schisandrin A, R(+)-
457	Schisandrin B, S(−)-	458	Schisantherin A	459	Securinine
460	Sedanolide	461	Silybine	462	Silymarin
463	Sinomenine	464	Solanesol	465	Vindoline
466	Vinorelbine base	467	Yangonin	468	Bergapten
469	Betulin	470	Corynanthine	471	Cytisine, (−)-
472	Spartein sulfate (−)-	473	Brassinin	474	Dihydrorobinetin
475	Flavanomarein	476	Lavendustin B	477	Evodiamine
478	Oxyacanthine sulfate	479	Galangine	480	Lavendustin A
481	Dihydroxy-4′-methoxy-chalcone, 2′,6′-	482	Gelsemine HCl	483	Hydrocotarnine HBr
484	Senecionine	485	Seneciphylline	486	Dihydrolysergol, 9,10-
487	Amphotericin B	488	Amygdalin	489	Anisodamine
490	Aphidicolin	491	Arbutin	492	Sclerotiorin
493	Bleomycin	494	Chartreusin	495	Chlorogenic acid
496	Geraldol	497	Coumestrol	498	Diindolylmethane
499	Ferulic acid	500	Geniposide	501	Hirsutine
502	Indole-3-carbinol

3.3. Molecular docking

In addition, structure-based strategies (pharmacophore mapping, homology modeling, docking) have been conducted for the prediction of the lead compound. Due to the lack of crystallographic 3D structures at a high resolution of targeting protein, few studies were carried out in the past. Ounissi et al. (2018) reported that molecular dynamics simulation was used to identify potential COX-2 inhibitors by 3D-QASR in the Chinese Universal Natural Product Database. Another publication also indicated that the COX-2 inhibitor was screened by 3D-QASR in benzopyran class (Yadav et al., 2017).

For further screening, we used MOE to what the COX-2 protein docks with virtual screened chemicals. Our results indicated that the COX-2 binding pocket successfully docked with osthole, kavain, vanillyl acetone, myristicin, and psoralen, and stabilized in electrostatic, van der Waals forces, and hydrogen interaction (Table 4, Figs. 6 and 7).

FIG. 6.

Three-dimensional view of the binding interaction of five virtual hits with COX-2. (A) Osthole, (B) psoralen, (C) kavain, (D) vanillyl acetone, and (E) myristicin. COX-2, cyclooxygenase-2.

FIG. 7.

Two-dimensional view of the binding interaction of five virtual hits with COX-2. (A) Osthole, (B) psoralen, (C) kavain, (D) vanillyl acetone, and (E) myristicin.

Table 4.

Chemicals Identified as COX-2 Inhibitor by FS-Random Forest (Feature Selection-Random Forest) Model and Molecular Docking in Nature Product Library

Chemical		S (kcal/mol)	RMSD (Å)
1	Kavain	−6.7255	1.7582
2	Osthole	−7.0467	1.9553
3	Vanillyl acetone	−6.2088	1.9952
4	Psoralen	−6.4018	1.5306
5	Myristicin	−5.8818	1.7615

RMSD, root mean square deviation.

3.4. Validation of inhibition activity in mRNA and protein level in HepG2 cell

As the selected library presented with the value of the half-maximal inhibitory concentration (IC₅₀), whether the chemical has cytotoxicity is unknown. To do this, HepG2 cells were treated with the five hit compounds, and LC₅₀ value of osthole, psoralen, and kavain was 120, 525, and 547 μM; furthermore, the lethal concentration 50 (LC₅₀) value of vanillyl acetone and myristicin was higher than 1000 μM (Fig. 8).

FIG. 8.

The cell viability was determined in HepG2 cell in a dose-dependent manner after the five hits' treatment. (A) Osthole, (B) psoralen, (C) kavain, (D) vanillyl acetone, and (E) myristicin.

To investigate whether screened chemicals from the natural product library suppress COX-2 expression in mRNA level, we performed real-time quantification in HepG.2 cells treated for 24 hours with the five hit compounds, but only osthole, psoralen, and kavain exhibited a decrease in COX-2 mRNA expression (Fig. 9). To investigate whether selected chemicals from the natural product library suppress COX-2 expression in protein level, we performed Western blotting in HepG2 cells treated for 24hr with the five hit compounds; only osthole, psoralen, and kavain inhibited COX-2 protein expression, yet myristicin elevated inversely (Fig. 10).

FIG. 9.

The mRNA level was determined by real-time PCR in HepG2 cell in a dose-dependent manner after the five hits' treatment. (A) Osthole, (B) psoralen, (C) kavain, (D) vanillyl acetone, and (E) myristicin.

FIG. 10.

The protein level was determined by Western blotting in HepG2 cell in a dose-dependent manner after the five hits' treatment. (A) Osthole, (B) psoralen, (C) kavain, (D) vanillyl acetone, and (E) myristicin.

Conclusion

In this work, a combination of RF and SOM neural network has been used to develop QSAR models that could discriminate COX-2 inhibitors from high and low activity, adopted in the natural product library to screen COX-2 inhibitors. According to the selected chemicals by virtual screening, we further investigated the interaction between COX-2 protein structure and chemical by molecular docking analysis in MOE. Through the above two steps, we predicted that osthole, kavain, vanillyl acetone, myristicin, and psoralen have COX inhibitor activity, however, only osthole, psoralen, and kavain showed COX inhibitor activity in mRNA and protein levels in HepG2 cells.

Footnotes

Acknowledgment

L.P. was supported by the Natural Science Foundation of Hainan Province (No. 817143).

Author Disclosure Statement

The authors declare they have no conflicting financial interests.

References

Aparoy

, Reddy

K.K.

, Kalangi

S.K.

, et al. 2010. Pharmacophore modeling and virtual screening for designing potential 5-lipoxygenase inhibitors. Bioorg. Med. Chem. Lett. 20, 1013–1018.

Breiman

2001. Random forests. Machine Learning. 45, 199–228.

Dong

, Cao

D.S.

, Miao

H.Y.

, et al. 2015. ChemDes: An integrated web-based platform for molecular descriptor and fingerprint computation. J. Cheminfo. 7, 60.

Franke

, Schwarz

, Mu

, et al. 2007. Identification of natural-product-derived inhibitors of 5-lipoxygenase activity by ligand-based virtual screening. J. Med. Chem. 50, 2640–2646.

Hegazy

G.H.

, and Ali

H.I.

2012. Design, synthesis, biological evaluation, and comparative Cox1 ACCEPTED MANUSCRIPT 39 and Cox2 docking of p-substituted benzylidenamino phenyl esters of ibuprofenic and mefenamic acids. Bioorganic Med. Chem. 20, 1259–1270.

Jović

, Brkić

, and Bogunović

2015. A review of feature selection methods with applications, 1200–1205. In 2015 38th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO). Opatija.

Kauffman

G.W.

, and Jurs

P.C.

2010. QSAR and k-nearest neighbor classification analysis of selective cyclooxygenase-2 inhibitors using topologically-based numerical descriptors. J. Chem. Info. Comp. Sci. 41, 1553–1560.

Kingston

D.G.I.

2011. Modern natural products drug discovery and its relevance to biodiversity conservation. J. Nat. Prod. 74, 496–511.

Lavecchia

2015. Machine-learning approaches in drug discovery: Methods and ap-plications. Drug Discov. Today. 20, 318–331.

10.

Limongelli

, Bonomi

, Marinelli

, et al. 2010. Molecular basis of cyclooxygenase enzymes (COXs) selective inhibition. Proc. Natl. Acad. Sci. U. S. A. 107, 5411–5416.

11.

Micklewright

, Lane

, Linley

, et al. 2003. Review article: NSAIDs, gastroprotection and cyclo-oxygenase-II-selective inhibitors. Aliment. Pharmacol. Ther. 17, 321–332.

12.

Mills

2006. ChemDraw Ultra 10.0 CambridgeSoft, 100 Cambridge Park Drive, Cambridge, MA 02140. www.cambridgesoft.com. Commercial Price: $1910 for download, $2150 for CD-ROM; Academic Price: $710 for download, $800 for CD-ROM. J. Am. Chem. Soc. 128, 13649–13650.

13.

Ounissi

, Kameli

, Tigrine

, et al. 2018. Computer-aided identification of lead compounds as cycooxygenase-2 inhibitors using virtual screening and molecular dynamic simulatin. Comput. Biol. Chem. 77, 1–16.

14.

Vesanto

, and Alhoniemi

2000. Clustering of the self-organizing map. IEEE Trans. Neural Netw. 11, 586–600.

15.

Yadav

D.K.

, Saloni Sharma

, Misra

, et al. 2017. Studies of the benzopyran class of selective COX-2 inhibitors using 3D-QSAR and molecular docking. Arch. Pharm. Res. DOI: 10.1007/s12272-017-0945-7

16.

Zheng

, Zhang

, Zhu

, et al. 2006. Essential structural profile of a dual functional inhibitor against cyclooxygenase-2 (COX-2) and 5-lipoxygenase (5-LOX): Molecular docking and 3D-QSAR analyses on DHDMBF analogues. Bioorg. Med. Chem. 14, 3428–3437.