Abstract
SARS-CoV-2, the virus that causes COVID-19, is a current concern for
people worldwide. The virus has recently spread worldwide and is out of
control in several countries, putting the outbreak into a terrifying
phase. Machine learning with transcriptome analysis has advanced in
recent years. Its outstanding performance in several fields has emerged
as a potential option to find out how SARS-CoV-2 is related to other
diseases. Idiopathic pulmonary fibrosis (IPF) disease is caused by
long-term lung injury, a risk factor for SARS-CoV-2. In this article,
we used a variety of combinatorial statistical approaches, machine
learning, and bioinformatics tools to investigate how the SARS-CoV-2
affects IPF patients’ complexity. For this study, we employed two
RNA-seq datasets. The unique contributions include common genes
identification to identify shared pathways and drug targets, PPI
network to identify hub-genes and basic modules, and the interaction of
transcription factors (TFs) genes and TFs–miRNAs with common
differentially expressed genes also placed on the datasets.
Furthermore, we used gene ontology and molecular pathway analysis to do
functional analysis and discovered that IPF patients have certain
standard connections with the SARS-CoV-2 virus. A detailed
investigation was carried out to recommend therapeutic compounds for
IPF patients affected by the SARS-CoV-2 virus.
Keywords: SARS-CoV-2, COVID-19, machine learning, idiopathic pulmonary
fibrosis, gene ontology, differentially expressed genes
Introduction
Coronaviruses have various variants that can infect humans and animals
[[40]1]. The variants of this virus are responsible for various
diseases, ranging from common fever and cold cough to more serious
illnesses such as Severe Acute Respiratory Syndrome (SARS) and Middle
East Respiratory Syndrome (MERS) [[41]2]. Severe Acute Respiratory
Syndrome Coronavirus 2 (SARS-CoV-2) is a new type of coronavirus, which
got a lot of attention at the end of 2019 because it was a new variant
of coronavirus that had never been observed in humans previously.
Coronavirus Disease 2019 (COVID-19) is the name of the new coronavirus,
which was first discovered in Wuhan, China, in December 2019 [[42]3].
Chinese officials reported 44 instances of pneumonia with unknown
causes to the World Health Organization (WHO) between 31 December 2019,
and 3 January 2020 [[43]4]. The first fatality of COVID-19 occurred in
Wuhan on 9 January 2020, while the first death outside of China
occurred in the Philippines on 1 February 2020. Within a few days, the
disease had spread worldwide and was out of control in many nations
[[44]4]. On 30 January 2020, the WHO designated the virus as a Public
Health Emergency (PHE) of worldwide concern [[45]5]. This virus was
declared a pandemic by the same organization on 11 March 2020, after a
total of 4500 deaths were reported in 30 countries and territories
throughout the world [[46]5]. Italy surpassed China, with the highest
reported death cases of this virus reported on 19 March 2020 [[47]4].
The USA has surpassed both China and Italy as the country with the
highest confirmed virus cases on 26 March 2020 [[48]4]. On a global
basis, the bloodiest week was 13–19 April 2020, when nearly 7460 deaths
were officially reported each day by this virus. The pandemic’s
epicenter migrated to Latin America and the Caribbean in June 2020.
Between 15 July 2020 and 15 August 2020, the region had an average of
almost 2500 deaths per day. With over 78 000 cases on 30 August 2020,
India surpassed the US record for the highest cases in a single day,
and a second wave hit India on 9 April 2021. There were 281 808 270
confirmed cases from December 2020 to December 2021, with 5 411 75
deaths by this virus [[49]6]. On 26 November 2021, WHO designated a new
variant (B.1.1.529) of SARS-CoV-2 named Omicron in South Africa. On 26
November, WHO designated Omicron as a variant of concern. The first
COVID-19 case associated with the Omicron variant was reported in the
USA on 1 December 2021, and at least one Omicron variant had been
detected in 22 states as of 8 December 2021. Recently, this new
variation of this virus has spread worldwide and is out of control in
several countries, putting the outbreak into a terrifying phase.
SARS-CoV-2 is a single-stranded RNA virus that is positive in a sense.
The Spike (S), envelope (E), membrane (M), and nucleocapsid (N)
proteins are the four proteins found in SARS-CoV-2. Spike proteins are
responsible for attaching to a host cell’s membrane. Idiopathic
pulmonary fibrosis (IPF) disease is a long illness marked by the
thickening and stiffening of lung tissue associated with scar tissue
formation [[50]7]. In this condition, the sponge or meaty section of
the lung becomes scarred or fibrotic. It is a slow-progressing, highly
fatal disease that affects roughly 80% of people within 3–5 years of
diagnosis [[51]7]. Pulmonary fibrosis affects people in different ways.
Various common, easily curable diseases might cause similar symptoms.
Shortness of breath and a persistent dry, hacking cough are the most
common indications and symptoms of IPF. Many impacted people also
notice a decrease in appetite and weight loss over time. Due to a lack
of oxygen, some people with IPF acquire enlarged, rounded tips on their
fingers and toes (clubbing) [[52]8]. IPF’s cause is not understood. The
following are some of the most common risk factors for IPF: Almost all
patients with IPF are over 50 years. Genetics, up to 20% of patients
with IPF have another family member who suffers from the condition.
Approximately 75% of people with IPF smoke now or have in the past.
Gastroesophageal reflux or heartburn affects about 75% of people with
IPF. Male patients account for roughly 65% of IPF patients [[53]9].
Radiation treatments to the chest or the use of certain chemotherapy
medications have been shown to enhance the risk of pulmonary fibrosis
[[54]10, [55]11]. SARS-CoV-2 contains spike protein, which has a
greater interaction with ACE2, and IPF patients have a lot of this
enzyme, confirming IPF as a risk factor for this disease [[56]12,
[57]13]. These investigations have revealed several linkages between
IPF and COVID-19, which raises concerns.
Contributions
In this article, we used a variety of combinatorial statistical
approaches, machine learning algorithms, and bioinformatics tools to
investigate how the SARS-CoV-2 virus affects IPF patients’ complexity.
The following are the main contributions of this article:
* the experiments have been conducted using a real-time dataset. We
have observed common gene identification by machine learning
algorithms and various bioinformatics analyses to identify shared
pathways and drug targets;
* the Protein-Protein Interaction (PPI) network was examined to
discover hub-genes and modules. The interactions of transcription
factors (TFs) genes and TFs-miRNAs with common differentially
expressed genes (DEGs) were also discovered. Furthermore, we used
gene ontology (GO) analyses and molecular pathway analyses to do
functional analysis and discovered that IPF patients have certain
common connections with SARS-CoV-2 infection;
* a comprehensive analysis has been conducted to suggest drug
molecules for IPF patients with SARS-CoV-2 infections. In the
context of molecular-based knowledge and several pathway-based
analyses, which illustrate the utility of the biological system for
both SARS-CoV-2 and IPF; and
* finally, the current challenges and future research directions of
integration and interplay between machine learning and
bioinformatics have been discussed.
The remainder of this study is organized in the following manner. The
‘Materials and methods’ section begins with a full description of the
dataset with preprocessing and an overview of selected methodology. The
‘Result analysis’ section discusses the evaluation and interpretation
of experimental outcomes for these methodologies. In addition,
‘Discussion’ section contains a lengthy explanation and discusses some
application areas for scientific society. Finally, ‘Conclusions’
section contains an overview of the findings and possible future
directions.
Materials and methods
In this section, we have thoroughly detailed the overview of the
analysis, including the dataset transformation process and various
transcriptome analyses.
Overview of approach
We applied machine learning and transcriptomic analysis to identify
shared associations between SARS-CoV-2 and IPF by employing selected
datasets shown in the block diagram in [58]Fig. 1. The machine learning
approaches have been used to identify common DEGs of the selected
datasets. Furthermore, these shared or common DEGs were used to
construct gene–disease association networks, identify GO, pathways, PPI
network, hub-genes, transcription factor (TF)–gene, TF–miRNA, and
identify candidate drugs.
Figure 1:
[59]Figure 1:
[60]Open in a new tab
The complete workflow for the current investigation. Two types of
samples (control cells, affected cells) were collected from
SARS-CoV-2-infected lung epithelial cells and both are included in the
[61]GSE147507 dataset. The [62]GSE52463 dataset contains IPF-affected
lung samples. Common DEGs were identified from both the datasets using
machine learning technique. From the common DEGs, GO identification,
pathway analysis, PPIs network, TF–gene analysis, TF–miRNA analysis,
and hub-gene identification were designed and based on those analysis
drug molecule identification was performed.
Dataset analysis
This section has performed a series of operations on the dataset
without changing its properties. Also, we have thoroughly explained the
overview of the selected dataset.
Dataset description
We have identified common genetic interrelationships between SARS-CoV-2
and IPF using Ribonucleic Acid Sequencing (RNA-Seq) datasets from the
Gene Expression Omnibus (GEO) collection of the National Center for
Biotechnology Information (NCBI) directory [[63]14, [64]15]. The
transcriptional responses to SARS-CoV-2 infection are contained in the
SARS-CoV-2 dataset with GEO accession ID [65]GSE147507 and GEO platform
ID [66]GPL18573. In contrast, the transcriptome analysis reveals
differential splicing events in IPF lung tissue that are contained in
the IPF dataset GEO accession [67]GSE52463 and GEO platform ID
[68]GPL11154 [[69]16]. SARS-CoV-2-affected Lung Epithelial Cell (LECs)
are found in the [70]GSE147507, while IPF-affected lung tissues are
found in the [71]GSE52463 dataset. The [72]GSE147507 dataset contains
two types of samples (control and SARS-CoV-2-affected cells) taken from
SARS-CoV-2-affected LECs, while the [73]GSE52463 dataset has two types
of samples (control and IPF-affected cells). Metadata and count data
are also included in both databases. The RNA sequence was extracted
from the [74]GSE147507 dataset using high-throughput sequencing
technologies on the Illumina NextSeq 500 (Homo sapiens) platform
[[75]17]. The IPF dataset, on the other hand, comprises mRNA sequencing
of eight IPF-affected lung tissues and seven control lung tissue
samples, all of which were sequenced on the Illumina Hi-Seq 2000
(H.sapiens) platform utilizing high-throughput sequencing technology
[[76]18]. [77]Table 1 lists the datasets used in this study and their
geo-features and sequencing methods.
Table 1:
Contents of the datasets
Properties SARS-CoV-2 IPF
GEO Accession [78]GSE147507 [79]GSE52463
GEO Platform [80]GPL18573 [81]GPL11154
Organisms Homo sapiens Homo sapiens
Assay type RNA-Seq RNA-Seq
Type of the datasets Transcriptional response to SARS-CoV-2 infection
In IPF lung tissue, transcriptome analysis indicates distinct splicing
events.
Instrument Illumina NextSeq 500 Illumina HiSeq 2000
Total GEO samples 110 15
Experiment type High-throughput sequencing for expression profiling
High throughput sequencing for expression profiling
[82]Open in a new tab
Data preparation
To achieve optimal performance, it is necessary to clean and prepare
the dataset before applying machine learning methods. Data preparation
is generally done by removing unnecessary features, checking the
variation of independent features, converting non-numerical features,
removing outliers, and replacing missing values if they exist. The two
fundamental steps apply during the data preparation process. The first
is data preprocessing, and the second is the data transformation step.
Data preprocessing
This dataset originates from multiple heterogeneous sources. Due to its
vast size, this dataset is highly susceptible to missing and noisy
data. This section discusses the essential steps in data preprocessing:
data cleaning and data integration.
* Data Cleaning: First, we applied various techniques to remove noise
and clean inconsistencies in the metadata and countdata from both
datasets. For example, Rosner’s test for outliers checking and the
predictive mean matching method for imputing missing values. Then,
to apply machine learning techniques, we converted the qualitative
values into quantitative values by applying various techniques
(e.g. Biobase (version 2.30.0), GEOquery (version 2.40.0), limma
(version 3.26.8), and Bioconductor) packages of the R programming
language, which is a free, open-source, and open-development
software project for the analysis and comprehension of genomic
data.
* Data Integration: To improve the accuracy, the data integration
technique helped us reduce and avoid redundancies in the resulting
dataset. This dataset originates from multiple heterogeneous
sources. So, it is essential to check both datasets for redundancy
and correlation analysis. This analysis has measured how strongly
one feature implies the other. [83]Figure 2a and b shows the
correlation between different features for the two datasets,
[84]GSE147507 and [85]GSE52463, respectively. For our analysis, we
have evaluated the correlation between all the features using the
following Pearson’s product-moment coefficient equation.
[MATH:
r=∑
i=1n<
mo> xi-x↼yi-y↼∑i
=1n <
/mo>xi-x↼2∑i=1<
mi>n yi-y↼2
:MATH]
Figure 2:
[86]Figure 2:
[87]Open in a new tab
The correlation analysis between different features for the two
datasets (a) [88]GSE147507 and (b) [89]GSE52463. This analysis has
measured how strongly one feature implies the other.
where ā is the meaning of x variable and
[MATH: y‾ :MATH]
is the meaning of y variable,
[MATH: xi
:MATH]
and
[MATH: yi
:MATH]
are values in tuple i.
Data transformation
We applied this processing step to achieve more efficient resulting
processes and easily understand the patterns. Some selected features
have larger values than others, which leads to incorrect performance.
We have implemented these strategies to scale the selected feature
values within a range between [0.0] and [1.0] without changing the
characteristics of the data.
N=(X−Xmin)/(Xmax−Xmin)
where N is the output normalized values, X is an original value and
Xmax and Xmin is the maximum and minimum values of the feature,
respectively.
As shown in the following equation, a technique called minimum–maximum
normalization has been used to scale the selected feature values within
the range. We have also evaluated the density plot for both datasets.
The density plot shows the smooth distribution of the points along the
numeric axis. The peaks of the density plot are at the locations where
there is the highest concentration of points. [90]Figure 3a and b shows
density plots for the two datasets, [91]GSE147507 and [92]GSE52463,
respectively.
Figure 3:
[93]Figure 3:
[94]Open in a new tab
The density plots of the two datasets (a) [95]GSE147507 and (b)
[96]GSE52463. The density plot shows the smooth distribution of the
points along the numeric axis. The peaks of the density plot are at the
locations where there is the highest concentration of points.
Spotting DEGs and shared DEGs between SARS-CoV-2 and IPF
A gene is differentially expressed when a statistical discrepancy
exists between several test settings during the transcription phase
[[97]19]. The major purpose of this study is to identify DEGs that are
shared between the [98]GSE147507 and [99]GSE52463 datasets. The DESeq2
and lima packages of the R programming language were used to access
data generated by microarray analysis. DEGs from both datasets were
identified using a machine learning method. Listing 1 shows the applied
procedure of the machine learning algorithm to identify DEGs from both
datasets. Across all datasets, significant DEGs were identified using
cutoff criteria (P-value < 0.05 and |log Fc| ≥ 1.00). The shared DEGs
of the [100]GSE147507 and [101]GSE52463 datasets were found using the
online VENN analysis platform JVENN tool [[102]20].
Listing 1. The procedure of the machine learning algorithm to identify
DEGs from both datasets.
1 Input: RNA-Seq dataset
2 Output: Identification of differentially expressed genes (DEGs)
3
4 outputFileName=open( “result.csv” , ”a” )
5
6 datasetName=os.listdir(folderPath)
7 for i in range(len(datasetName)):
8 fileName=open(“GSE”+str(name[i])+”.csv”)
9 fileName.next()
10 for dataset in fileName:
11 #Extract countdata and store in a matrix
12 datasetName=os.listdir(folderPath)
13 countData = read.csv((“GSE”+str(name[i])+ ”filtered_countdata.csv
”):
14 countDataFrame = data.frame(countData)
15 countDataFrameRound=mutate(across(where(is. numeric ),round
, 3))
16 #Extract metadata and store in a matrix
17 metaData = read.csv((“GSE”+str(name[i])+ ”filtered_metadata.csv ”):
18 countDataFrame = data.frame(metaData)
19 #Analyze count data using DESEQ2
20 applyDESeq = DESeqDataSetFromMatrix(countData=countDataFrameRound,
21 colData=metaDataFrame, design=∼treatment, tidy= TRUE )
22 applyDESeq = DESeq(applyDESeq)
23 result = results(applyDESeq)
24
25 #Result Analysis
26 #Check and Omit the null value
27 checkNull = is.na(result)
28 resultsOmitNa = na.omit(result)
29
30 #Count the up regulated gene
31 resultOmitNaFilterUp = filter(resultOmitNa, log2FoldChange
32 > 1 &Padj < 0.05)
33 #Count the down regulated gene
34 resultOmitNaFilterDown = filter(resultOmitNa, log2FoldChange
35 <-1 & P adj < 0.05)
36 #ABS logFC value and setup cuttoff criteria for P adj value
37 resultFinal = filter(resultOmitNa, abs(log2FoldChange)
38 > 1 &Padj < 0.05)
39 outputFileName.write(resultFinal)
40 outputFileName.close()
Identifying of GO and molecular pathway
Enrichment analysis of gene set is a technique for identifying DEGs
linked to a biological process or molecular function [[103]21]. GO is a
classification system that divides genes into biological mechanisms,
molecular functions, and cellular components [[104]22]. The purpose of
analyzing GO concepts is to understand the molecular activity, cellular
structure, and the position in the cell where genes fulfill their
functions [[105]23]. We used four databases to find common molecular
pathways in IPF and COVID-19: Kyoto Encyclopedia of Genes and Genomes
(KEGG) [[106]24], Wiki Pathways [[107]24], Reactome [[108]25], and
BioCarta [[109]25]. Various gene annotations may be found in the KEGG,
which is commonly used to characterize metabolic pathways. A web-based
platform Enrichr has been used to obtain GO, and molecular pathways for
the common genes mentioned earlier in this research [[110]26, [111]27].
To derive GO and molecular pathways, we utilized 20 sorted genes.
Analysis of PPI network
The role of PPIs in cellular biology is projected to be a major focus
of research, and it serves as a requirement for system biology
[[112]28]. Proteins finish their journey within a cell with a
comparable protein affiliation established by a PPI network, indicating
the protein processes. Proteins interact with other proteins to carry
out their activities inside cells, and the information created by a PPI
network informs individuals about the protein’s function [[113]29]. We
built the PPI network of DEGs proteins using the STRING resource to
exchange activity and physical linkages between IPF and COVID-19
[[114]30]. The STRING generates experimental and predicted outcomes
based on the data and the interaction generated by the online tool,
which is determined by 3D structures, accessory data, and confidence
scores [[115]31]. The confidence score was set using the STRING
platform that was different categorized confidence scores (low, medium,
and high). We have been worked on the PPI network with a medium
confidence score (0.400). We get the exact information, using the
network type “full string network” (the edges indicate both functional
and physical protein resources) and a selected number of 10
interactors. Then, we consume our PPI network into Cytoscape (version
3.7.1) for visual representation and further PPI network experimental
studies. And with that the purpose of identifying hub-genes, the
obtained PPIs are analyzed through Cytoscape. Cytoscape is an
open-source network visualization framework that serves as a versatile
method for combining several datasets to optimize efficiency for
various interactions such as protein–protein interactions, genetic
interactions, and protein–DNA interactions, among others [[116]32,
[117]33].
Identifying of hub-genes and module analysis
The PPI networks are nodes, edges, and connections, with hub-genes
being the most entangled nodes. The PPI networks are used to identify
hub-genes. Hub-genes provide dense areas identified as important parts
of the PPIs network. The hub-genes for the associated PPI networks are
indicated by CytoHubba, a Cytoscape application plugin [[118]34].
CytoHubba is the most popular Cytoscape hub-genes identification plugin
for its user-friendly interface. CytoHubba has 20 different methods for
topological analysis (e.g. MCC, Degree, DMNC, MNC, EPC, Bottleneck,
etc.). The degree analysis method was employed to find the hub-genes
for this study. Because the degree method facilitates analysis by
suggesting large, closely compacted modules in the PPI network, it is
employed instead of another approach [[119]35]. The Molecular Complex
Detection (MCODE) plugin in the Cytoscape software is utilized to
locate the most profound modules in the PPIs network [[120]36]. The
MCODE method is based on a graph-theoretic clustering algorithm that
detects densely connected regions in large protein–protein interaction
networks that may represent molecular complexes [[121]36]. The method
has the advantage over other graph clustering methods of having a
directed mode that allows fine-tuning of clusters of interest without
considering the rest of the network and allows examination of cluster
inter-connectivity, which is relevant for protein networks.
Furthermore, the method is not affected by the known high rate of false
positives in data from high-throughput interaction techniques
[[122]37]. Moreover, the method is relatively easy to implement and,
since it is local density based, has the advantages of both a directed
mode and a complex connectivity mode. The MCODE method has also been
employed in the PPIs network to locate highly bound areas in the
molecular complexes.
TF–gene analysis
TFs bind to individual genomes and regulate their levels of expression.
As a result, it is required for molecular recognition [[123]38]. In all
species, TFs control gene expression and play a critical role in
transcription. TFs play an important role in a variety of biological
processes, including cell cycle regulation and development. TF–gene
linkage with the newly discovered top 12 common DEGs among 90 DEGs was
used to investigate the effects of TF–genes on functional pathways and
genomic levels. By using the Network Analyst tool to find topologically
relevant TFs from the ENCODE database, which was used in the TF–gene
interaction network [[124]39–41], we were able to exploit TF–gene
interactions with previously established common genes. Network Analyst
is a web-based tool for doing transcriptional research and
meta-analysis on various species, including humans [[125]42, [126]43].
The TF–gene interaction network has made up of 190 nodes and 301 edges.
Moreover, the network has 12 DEGs and 178 TF–genes, where HSPB6 is
regulated by 85 TF–genes, EPAS1 is regulated by 68 TF–genes, and FCGR2A
is regulated by 37 TF–genes according to their degree value. These 178
TF–genes are regulated by more than one common DEG, which indicates
high interaction of the TF–genes with common DEGs.
TF–miRNA interaction with the common DEGs
The miRNAs are short non-RNAs that are expressed by RNA polymerase II
and then regulated by a shared biogenic pathway in a step-by-step
method. Using a combination of experimental and computational
techniques, miRNAs have been discovered in a variety of species. By
binding to the 3′-untranslated, miRNA regulates gene expression at the
post-transcriptional stage. The RegNetwork database was utilized to
collect TF–miRNA coregulatory interactions, which helps to identify the
miRNAs and regulatory TF–genes that regulate DEGs of interest at the
transcriptional and post-transcriptional phases [[127]43]. We found
miRNAs that interact with common DEGs and then utilized the Network
Analyst tool to analyze how they interact. With this platform,
researchers can find complex datasets and determine biological traits
and functions [[128]44]. The network of miRNA–gene interactions was
examined using Cytoscape software. By classifying top miRNAs to higher
levels, this software aids researchers in determining biological roles
and features. The TF–miRNA coregulatory network has 191 nodes and 216
edges. According to research, DEGs engage with 87 miRNAs and 93
TF–genes.
Candidate drugs identification
Predicting PDI or drug molecule recognition is important for this
research. We identified a therapeutic molecule based on the common DEGs
of SARS-CoV-2 and IPF using the Enrichr tool and DSigDB database. There
are 22 527 gene sets in the drug signatures database. To acquire access
to the DSigDB database, the Enrichr platform is employed [[129]45,
[130]46]. Enrichr is a well-known web portal with many gene-set
libraries that may be used to look into gene-set enrichment on a
genome-wide scale [[131]26].
Result analysis
The overall performance of the analysis is discussed in this section.
Beginning with a discussion of DEGs and mutual DEG identification, the
article progresses to a description of the candidate drug
identification procedure.
DEGs and mutual DEGs identification
We investigated the interrelationships and implications of disrupted
genes that activate COVID-19 and IPF using the NCBI’s human RNA-seq and
microarray datasets. The [132]GSE147507 dataset determines DEGs for
SARS-CoV-2, and its GEO platform identifier is [133]GPL18573. There are
926 upregulated and 799 downregulated genes in the [134]GSE147507
dataset, resulting in 1725 DEGs. In the [135]GSE52463 dataset, which
has the GEO platform identifier [136]GPL11154, we discovered a total of
1008 DEGs, with 669 upregulated and 339 downregulated genes. The
quantitative measurement of the selected datasets is shown in
[137]Table 2. After cross-comparative analysis using JVENN, a
trustworthy web platform for Venn analysis, we discovered 90 similar
DEGs from the [138]GSE147507 and [139]GSE52463 datasets. Twenty common
DEGs were chosen for further study from 90 common DEGs based on the
P-value (MDK, HP, HSPB6, CHIT1, TNFAIP6, EPAS1, MMP1, CCL18, CXCL6,
CCL11, IL1RN, LAMP3, CD207, ARRB1, RNASE2, LILRA1, FCGR2A, STAT4, CD69,
and SAMSN1). Additional study has been conducted using these 20
frequent DEGs. [140]Figure 4 depicts the common DEGs as a Venn diagram,
with 90 genes discovered to be shared in the [141]GSE147507 and
[142]GSE52463 datasets.
Table 2:
Quantitative measurements of the datasets used in this analysis
Properties [143]GSE147507 [144]GSE52463
Common gene analysis DESeq2 and the lima package DESeq2 and the lima
package
Cutoff criteria P < 0.05 and |log Fc| ≥ 1.0 P < 0.05 and |log Fc| ≥ 1.0
Total DEGs count 1725 genes 1008 genes
Upregulated DEGs count 926 genes 669 genes
Downregulated DEGs count 799 genes 339 genes
[145]Open in a new tab
Figure 4:
[146]Figure 4:
[147]Open in a new tab
Common DEGs representation through a Venn diagram. There are 90 genes
were found common from the 1635 DEGs of SARS-CoV-2 infection and 918
DEGs of IPF patients. The common DEGs were 3.4% among total 2553 DEGs.
GO and molecular pathway analysis
Enrichment analysis of gene sets is a technique for identifying DEGs
linked to a biological process or molecular function. For this study,
we looked at the most prevalent DEGs. GO processes are divided into
biological, cellular components, and molecular functions. [148]Table 3
shows the biological process connected to GO keyword identification
findings based on the combined score. [149]Table 4 shows the results of
the identification of molecular function-related GO keywords based on
the combined score. [150]Table 5 also shows the results of the cellular
component-related GO keywords identification based on the combined
score. The KEGG, Wiki Pathways, Reactome, and BioCarta have been used
to find the most impactful pathways of the shared DEGs between IPF and
SARS-CoV-2. [151]Tables 6, [152]7, [153]8, and [154]9 show the
essential pathways discovered in the datasets. The graphical view of GO
terms and pathways analysis are shown in [155]Figs. 5 and [156]6.
Table 3:
The combined score was used to identify biological process-related GO
keywords
Group GO ID GO pathways P-value Genes
GO biological process GO: 0006032 Chitin catabolic process 6.98E-03
CHIT1
GO: 0090240 Positive regulation of histone H4 acetylation 6.98E-03
ARRB1
GO: 0006030 Chitin metabolic process 6.98E-03 CHIT1
GO: 0072677 Eosinophil migration 2.59E-04 CCL11; CCL18
GO: 0048245 Eosinophil chemotaxis 2.59E-04 CCL11; CCL18
GO: 0070098 Chemokine-mediated signaling pathway 1.83E-05 CXCL6; CCL11;
CCL18
GO: 0030593 Neutrophil chemotaxis 1.94E-05 CXCL6; CCL11; CCL18
GO: 0002029 Desensitization of G-protein coupled receptor protein
signal 7.97E-03 ARRB1
GO: 0038114 Interleukin-21-mediated signaling pathway 7.97E-03 STAT4
GO: 0098757 Cellular response to interleukin-21 7.97E-03 STAT4
[157]Open in a new tab
Table 4:
The combined score was used to identify GO keywords linked to molecular
functions
Group GO ID GO pathways P-value Genes
GO molecular function GO: 0019966 Interleukin-1 binding 5.98E-03 IL1RN
GO: 0008009 Chemokine activity 1.26E-05 CXCL6; CCL11; CCL18
GO: 0004568 Chitinase activity 6.98E-03 CHIT1
GO: 0042379 Chemokine receptor binding 1.53E-05 CXCL6; CCL11; CCL18
GO: 0005537 Mannose binding 1.09E-02 CD207
GO: 0048020 CCR chemokine receptor bind 6.54E-04 CCL11; CCL18
GO: 0005041 Low-density lipoprotein receptor 1.29E-02 TNFAIP6
GO: 0005125 Cytokine activity 1.53E-05 CXCL6; IL1RN; CCL11;
GO: 0005149 Interleukin-1 receptor binding 1.49E-02 IL1RN
GO: 0005159 Binding of insulin-like growth factor receptors 1.49E-02
ARRB1
[158]Open in a new tab
Table 5:
The combined score was used to identify cellular component-related GO
keywords
Group GO ID GO pathways P-value Genes
GO cellular component GO: 1904724 Tertiary granule lumen 1.37E-03
CHIT1; TNFAIP6
GO: 0030669 Clathrin-coated endocytic vesicle membrane 3.25E-02 CD207
GO: 0045334 Clathrin-coated endocytic vesicle 4.88E-02 CD207
GO: 0070820 Tertiary granule 1.15E-02 CHIT1; TNFAIP6
GO: 0030659 Cytoplasmic vesicle membrane 5.27E-02 ARRB1
GO: 0035580 Specific granule lumen 6.02E-02 CHIT1
GO: 0031410 Cytoplasmic vesicle 1.92E-02 CD207; ARRB1
GO: 0005769 Early endosome 2.04E-02 LAMP3; CD207
GO: 0031901 Early endosome membrane 7.05E-02 CD207
GO: 0030665 Clathrin-coated vesicle membrane 7.79E-02 CD207
[159]Open in a new tab
Table 6:
Pathway analysis results in identification through KEGG using the
combined score
Database Pathways P-value Gene
KEGG IL-17 signaling pathway 1.05E-04 CXCL6; CCL11; MMP1
Chemokine signaling pathway 3.39E-05 CXCL6; CCL11; ARRB1; CCL18
Cytokine–cytokine receptor interaction 1.84E-04 CXCL6; IL1RN; CCL11;
CCL18
Rheumatoid arthritis 3.69E-03 CXCL6; MMP1
Asthma 3.06E-02 CCL11
Osteoclast differentiation 7.05E-03 FCGR2A; LILRA1
Relaxin signaling pathway 7.37E-03 MMP1; ARRB1
Bladder cancer 4.02E-02 MMP1
Hedgehog signaling pathway 4.59E-02 ARRB1
Amino sugar and nucleotide sugar metabolism 4.69E-02 CHIT1
[160]Open in a new tab
Figure 5:
[161]Figure 5:
[162]Open in a new tab
According to the combined score, (a) biological, (b) molecular
function, and (c) cellular component relevant GO keywords were
identified. The higher the enrichment score, the higher number of genes
are involved in a certain ontology.
Figure 6:
[163]Figure 6:
[164]Open in a new tab
The pathway analysis results were identified using (a) KEGG, (b) Wiki
Pathways, (c) Reactome, and (d) BioCarta. The results of the pathway
terms were identified through the combined score.
Table 7:
Pathway analysis results in identification through Wiki pathways using
the combined score
Database Pathways P-value Gene
Wiki Pathways Thymic Stromal Lymphopoietin Signaling Pathway 1.00E-03
CCL11; STAT4
Amplification and Expansion of Oncogenic Pathways as Metastatic Traits
1.69E-02 EPAS1
Matrix Metalloproteinases 2.95E-02 MMP1
Signal transduction through IL1R 3.25E-02 IL1RN
Type 2 papillary renal cell carcinoma 3.34E-02 EPAS1
Photodynamic therapy-induced NF-kB survival signaling 3.44E-02 MMP1
Bladder Cancer 3.92E-02 MMP1
Integrated Cancer Pathway 4.31E-02 MMP1
Hedgehog Signaling Pathway 4.31E-02 ARRB1
Hepatitis C and Hepatocellular Carcinoma 4.79E-02 MMP1
[165]Open in a new tab
Table 8:
Pathway analysis results in identification through Reactome using the
combined score
Database Pathways P-value Gene
Reactome PTK6 Expression 4.99E-03 EPAS1
Regulation of gene expression by Hypoxia-inducible Factor 9.96E-03
EPAS1
Chemokine receptors bind chemokines 1.42E-03 CXCL6; CCL11
Oxygen-dependent proline hydroxylation of Hypoxia-inducible Factor
Alpha 1.78E-02 EPAS1
Activation of SMO 1.78E-02 ARRB1
Regulation of Insulin-like Growth Factor transport and uptake by
Insulin-like Growth Factor Binding Proteins 2.08E-02 MMP1
NOTCH2 Activation and Transmission of Signal to the Nucleus 2.08E-02
MDK
Basigin interactions 2.47E-02 MMP1
Regulation of hypoxia-inducible Factor by oxygen 2.56E-02 EPAS1
Cellular response to hypoxia 2.57E-02 EPAS1
[166]Open in a new tab
Table 9:
Pathway analysis results in identification through BioCarta using the
combined score
Database Pathways P-value Gene
BioCarta Beta-arrest ins in GPCR Desensitization Pathway 3.54E-04
CCL11; ARRB1
NO2-dependent IL12 Pathway in NK cells Pathway 8.96E-03 STAT4
Role of Beta-arrestins in the activation and targeting of MAP kinases
Pathway 4.06E-04 CCL11; ARRB1
G-Protein Signaling Through Tubby Proteins Pathway 9.95E-03 CCL11
Roles of Beta-arrestins-dependent Recruitment of Src Kinases in GPCR
Signaling Pathway 5.23E-04 CCL11; ARRB1
Activation of PKC through G-protein coupled receptors Pathway 1.09E-02
CCL11
Visual Signal Transduction Pathway 1.29E-02 ARRB1
Attenuation of GPCR Signaling Pathway 1.29E-02 ARRB1
IL12- and Stat4-dependent Signaling Pathway in Th1 Development 1.49E-02
STAT4
Cystic fibrosis transmembrane conductance regulator (CFTR) and beta 2
adrenergic receptor (b2AR) 1.98E-02 CCL11
[167]Open in a new tab
Analysis of PPI network for the identification of hub-genes
The PPI network analysis is the most important element. This network
has conducted hub-gene recognition, module analysis, and drug
identification. In STRING, the specific DEGs have been provided as
input. The analysis file was re-imported into the Cytoscape software
for visualization. For the most frequent DEGs, a PPI network has been
created. Finally, the PPIs network results connect to therapeutic
compound suggestions, placing the PPIs analysis as the research’s
focus. [168]Figure 7 shows the PPI network with 60 nodes and 308 edges.
For SARS-CoV-2 and IPF, the PPI network was developed to discover
hub-genes and medicinal compounds.
Figure 7:
[169]Figure 7:
[170]Open in a new tab
A network of PPIs discovered common DEGs in two illnesses (SARS-CoV-2
and IPF). The orange nodes denote common DEGs, whereas the edges denote
the relationship between two genes. The network under investigation has
60 nodes and 308 edges.
Identification of hub-genes for therapeutic solutions and module analysis
CytoHubba, a Cytoscape software plugin, was used to track the hub-genes
from the PPIs network. The degree meaning of the hub-genes, which
represents the number of interactions between the genes in the PPI
network, has been categorized. Hub-genes are the bulk of interconnected
nodes in a PPI network. The topological analysis identified the top
five genes (AKT1, IL1B, CCL5, MMP9, and ARRB1) classified as hub-genes
based on their degree value. [171]Table 10 shows the results of the
topological analysis. These hub-genes could be exploited as biomarkers,
leading to new therapeutic approaches for the studied diseases. The
network has 50 nodes and 283 edges, and we utilized a degree-sorted
circle structure to lay it out. The network of hub-genes is depicted in
[172]Fig. 8, with the top five hub-genes AKT1, IL1B, CCL5, MMP9, and
ARRB1.
Table 10:
Exploration of topological results for the top five hub-genes
Hub gene Degree Stress Close ness Between ness Bottle neck Clustering
coefficient EcCentricity Radiality
AKT1 27 3322 42.25000 637.30186 26 0.25356 0.25000 4.47458
IL1B 26 2172 42.33333 475.08574 03 0.34154 0.33333 4.52542
CCL5 22 1216 38.25000 238.70899 14 0.35931 0.25000 4.23729
MMP9 22 1808 39.16667 322.49125 07 0.35498 0.33333 4.33898
ARRB1 19 1630 37.55000 291.37776 06 0.43865 0.25000 4.25424
[173]Open in a new tab
Figure 8:
[174]Figure 8:
[175]Open in a new tab
The PPIs network was used to find hub-genes. There are 50 nodes and 283
edges in the network. AKT1 and IL1B have degrees of 27 and 26,
respectively, according to topological analysis. CCL5, MMP9, and ARRB1
had degrees of 22, 22, and 19, respectively.
TF–gene analysis
The Network Analyst platform was used to investigate TF–gene
interactions. The common DEGs were used to examine the TF–gene network.
There are 190 nodes and 301 edges in the TF–gene network. Furthermore,
the network contains 12 DEGs and 178 TF–genes, with 85 TF–genes
regulating HSPB6, 68 TF–genes regulating EPAS1, and 37 TF–genes
regulating FCGR2A according to their degree value. These 178 TF–genes
are regulated by several common DEGs, indicating a high level of
interaction between the TF–genes and common DEGs. The TF–gene network
is shown in [176]Fig. 9.
Figure 9:
[177]Figure 9:
[178]Open in a new tab
The interaction of TF–genes with common DEGs is represented via a
network. The common genes are shown by the highlighted yellow color
node, while TF–genes are represented by the other nodes. There are 190
nodes and 301 edges in the network.
TF–miRNA analysis
The TF–miRNA coregulatory network was built using the Network Analyst
tool. Analyzing this TF–miRNA coregulatory network revealed the
connection of miRNAs and TFs with common DEGs. There are 191 nodes and
216 edges in this coregulatory network. DEGs interact with 87 miRNAs
and 93 TF–genes, according to this study. [179]Figure 10 shows the
TF–miRNA coregulatory network.
Figure 10:
[180]Figure 10:
[181]Open in a new tab
There are 93 TF–genes, 87 miRNAs, and 11 DEGs in the TF–miRNA network.
There are 191 nodes and 216 edges in the network. DEGs are represented
by blue nodes, while miRNA is represented by green nodes, and TF–genes
are represented by other nodes.
Candidate drugs identification and validation
Drug compounds for common DEGs have been discovered using the Enrichr
platform. Using the DSigDB database, we discovered 10 candidate
medicinal compounds. The top 10 chemical compounds have been extracted
based on the combined score of P-value and adjusted P-value. NICKEL
SULFATE CTD 00001417, Clonidine HL60 UP, and THYMOLPHTHALEIN CTD
00006891 are the three-drug compounds most genes interact with,
according to the data. These medicines are common pharmaceuticals for
COVID-19 and IPF since these signature drugs have been discovered for
common DEGs. [182]Table 11 displays the most efficient medications for
the most common DEGs from the DSigDB database.
Table 11:
The top 10 drug compounds suggested for common DEGs
Name of the drugs P-value Adjusted P-value Name of the genes
Nickel Sulfate CTD 00001417 1.37E-12 8.81E-10 CXCL6; IL1RN; CCL11;
TNFAIP6; EPAS1; MMP1; LAMP3; CD207; STAT4; CD69; SAMSN1
Clonidine HL60 UP 1.04E-06 3.36E-04 IL1RN; FCGR2A; RNASE2; SAMSN1
Thymolphthalein CTD 00006891 3.80E-04 1.01E-02 EPAS1; ARRB1
Peptidoglycan CTD 00006490 4.34E-04 1.07E-02 TNFAIP6; MMP1
Lithocholic acid HL60 UP 4.63E-04 1.10E-02 CD69; SAMSN1
Beclomethasone CTD 00005468 3.93E-05 3.21E-03 IL1RN; CCL11; RNASE2
Salmeterol CTD 00002421 4.92E-04 1.13E-02 CCL11; RNASE2
Mephentermine HL60 UP 4.48E-05 3.21E-03 IL1RN; EPAS1; CD69
Colchicine HL60 UP 8.09E-06 1.04E-03 IL1RN; FCGR2A; EPAS1; SAMSN1
Bromocriptine HL60 UP 6.94E-05 4.07E-03 FCGR2A; TNFAIP6; SAMSN1
[183]Open in a new tab
Computationally predicted results usually need experimental
verification, but it has more difficulty and limitations in practical
implementation. Thus, similar to Zhang et al. [[184]47], they found a
novel validation process for suggested drug compounds based on the
Receiver Operator Characteristic (ROC) curve. We tried to validate our
suggested drug compounds using the ROC curve mechanism. [185]Figure 11
shows the validation performance comparison between the top five
suggested drug compounds using the ROC curve. We considered the top
five suggested drug compounds, where Nickel Sulfate has a higher
validation accuracy than the others, according to the ROC curve. Other
suggested drug compounds, as shown in [186]Fig. 11, were also validated
using the same procedures, which is much more valuable to the medical
community.
Figure 11:
Figure 11:
[187]Open in a new tab
Performance comparison of the top five suggested drug compounds based
on the ROC curve. We considered the top five suggested drug compounds,
where Nickel Sulfate has a higher validation accuracy than the others,
according to the ROC curve.
Discussion
COVID-19 is more common in people who have lung disease. This study
contributes to the development of a bioinformatics and machine learning
model to identify the Genetic Effect of SARS-CoV-2- and IPF-affected
patients. Shortness of breath, cough, and chest pain are the most
typical symptoms of these two diseases. About 1725 and 1008 DEGs were
found in [188]GSE147507 and [189]GSE52463, respectively, using
bioinformatics-related techniques. Common DEGs between the
[190]GSE147507 and [191]GSE52463 datasets have been discovered for
better coordination. There is a total of 90 DEGs that have been
identified. Twenty common DEGs were chosen for further study from 90
common DEGs based on the P-value (MDK, HP, HSPB6, CHIT1, TNFAIP6,
EPAS1, MMP1, CCL18, CXCL6, CCL11, IL1RN, LAMP3, CD207, ARRB1, RNASE2,
LILRA1, FCGR2A, STAT4, CD69, and SAMSN1). The analysis of GO, KEGG,
Wiki Pathways, Reactome, BioCarta pathway analysis, PPIs, TF–gene,
TF–miRNA coregulatory network, and candidate drug detection has been
continued in the research project.
DEGs that have been identified as common have been used to find GO
words. GO keywords were identified using the combined score. Biological
process, molecular function analysis, and cellular component analysis
are the three categories of GO analysis [[192]48]. KEGG, Wiki Pathways,
Reactome, and BioCarta were used to identify pathway analysis results.
For the most prevalent DEGs, the KEGG pathway has been determined. KEGG
is a database that aids researchers in understanding the high-level
functions and utility of biological systems. Because hub-gene
recognition, module analysis, and drug identification are all strongly
dependent on the PPI network, it is the significant part of the
research. Common DEGs were also subjected to PPI analysis. The
identification of hub-genes in the PPI network was studied. The five
genes that have been highlighted are AKT1, IL1B, CCL5, MMP9, and ARRB1.
These five genes are classified as hub-genes based on their degree
value. The aim of concentrating on a small area is to suggest a more
effective medication component.
The interaction of TF–genes and miRNAs was investigated to identify
transcriptional and post-transcriptional regulators of common DEGs. The
specific DEGs have been used to investigate TF–gene interactions.
TF–genes act as regulators of gene expression, which can contribute to
cancer cell formation. About 85 TF–genes regulate HSPB6, 68 TF–genes
regulate EPAS1, and 37 TF–genes regulate FCGR2A according to their
degree value in the network, with 12 DEGs and 178 TF–genes. The
TF–miRNA coregulatory network depicts the interactions between miRNAs
and TF–genes tested for their ability to influence common DEGs. There
were 87 miRNAs and 93 TF–genes discovered. Several studies have found
evidence of altered miRNA expression in IPF samples, and members of the
miR-200 family play a significant role in IPF sample management
[[193]49]. Taz et al. [[194]50] investigated only 69 samples, whereas
we analyzed 110 SARS-CoV-2 samples. As a result, this research will
ideally integrate COVID-19 with IPF risk factor treatment. Chemical
testing can be used to verify the drugs’ efficacy.
In addition, we thoroughly discussed the application areas of our
research for the scientific society. First of all, researchers can use
the same approach to investigate the impact of SARS-CoV-2 on other
diseases. Also, if a new virus appears, our research will serve as a
useful starting point for further investigation. Furthermore, our
research suggests several viable drugs, so scientists will be able to
find a treatment for SARS-CoV-2 with more research. Finally, our
research is an example of a virus's genetic relationship with a certain
type of patient. So, researchers can use this methodology to figure out
the genetic relationships between different viruses and patients.
Conclusions
COVID-19 infections have been associated with a high-risk factor for
IPF patients. Shortness of breath, cough, and chest pain are the most
typical symptoms of these two diseases. We used machine learning and
bioinformatics analysis to summarize the relationships between these
two disease genes as part of our research. We analyzed DEGs from two
selected datasets, analyzed the results using shared gene
identification, and discovered SARS-CoV-2- and IPF-affected lung-cell
infection responses. As a consequence, we discovered 90 genes that are
linked across these datasets. These interconnected genes built the PPI
network, which identified the five most important hub-genes. In
addition, we looked at SARS-CoV-2 and IPF to see if they might predict
the outcomes of identifying infections of other diseases. The
therapeutic goals are logically presented because they are executed
from the discovery of hub-genes and could work as an effective
precursor to meanwhile licensed medications. We believe that the
biomarkers, pathways, and molecular markers we discovered will be
valuable in developing pharmacological therapies.
Declarations
Ethical Approval
Not applicable (there is no human-related data. So, ethical approval is
not taken from the external body of the committee).
Consent to Participate
Not applicable (there is no human-related data. So, consent is not
necessary to take from the participant).
Consent to Publish
Not applicable (there is no human-related data. So, consent to publish
is not necessary to take from the participant).
Funding
This work was supported by Researchers Supporting Project number
(RSP-2021/100), King Saud University, Riyadh, Saudi Arabia. This work
was supported in part by funding from the Natural Sciences and
Engineering Research Council of Canada (NSERC).
Acknowledgement