Abstract
Analysis of Pan-omics Data in Human Interactome Network (APODHIN) is a
platform for integrative analysis of transcriptomics, proteomics,
genomics, and metabolomics data for identification of key molecular
players and their interconnections exemplified in cancer scenario.
APODHIN works on a meta-interactome network consisting of human
protein–protein interactions (PPIs), miRNA-target gene regulatory
interactions, and transcription factor-target gene regulatory
relationships. In its first module, APODHIN maps proteins/genes/miRNAs
from different omics data in its meta-interactome network and extracts
the network of biomolecules that are differentially altered in the
given scenario. Using this context specific, filtered interaction
network, APODHIN identifies topologically important nodes (TINs)
implementing graph theory based network topology analysis and further
justifies their role via pathway and disease marker mapping. These TINs
could be used as prospective diagnostic and/or prognostic biomarkers
and/or potential therapeutic targets. In its second module, APODHIN
attempts to identify cross pathway regulatory and PPI links connecting
signaling proteins, transcription factors (TFs), and miRNAs to
metabolic enzymes via utilization of single-omics and/or pan-omics data
and implementation of mathematical modeling. Interconnections between
regulatory components such as signaling proteins/TFs/miRNAs and
metabolic pathways need to be elucidated more elaborately in order to
understand the role of oncogene and tumor suppressors in regulation of
metabolic reprogramming during cancer. APODHIN platform contains a web
server component where users can upload single/multi omics data to
identify TINs and cross-pathway links. Tabular, graphical and 3D
network representations of the identified TINs and cross-pathway links
are provided for better appreciation. Additionally, this platform also
provides few example data analysis of cancer specific, single and/or
multi omics dataset for cervical, ovarian, and breast cancers where
meta-interactome networks, TINs, and cross-pathway links are provided.
APODHIN platform is freely available at
[31]http://www.hpppi.iicb.res.in/APODHIN/home.html.
Keywords: meta-interactome, network analysis, pan-omics, multi-omics
analysis, pathway cross links
Introduction
Technological advances have made different types of omics data
accessible in large scale. Different types of omics data are outcomes
of profiling of different bio-entities, namely RNA (RNA
transcriptomics), miRNA (miRNA transcriptomics), proteins (proteomics,
phosphoproteomics), genes (genomics, epigenomics), metabolites
(metabolomics), lipids (lipidomics), and pharmacogenomics. These
bio-entities are functionally inter-related in a complex fashion.
Extrapolation from single omics data of one type of bio-entity fails to
provide the true biological status of various linked bio-entities
(e.g., RNA, protein, metabolites). Hence, to inquire the causative
phenomena underlying the genesis and progression of systemic/genetic
diseases, an integrative analysis considering the profiles of above
mentioned bio-entities appears as a requisite. Moreover, because of the
heterogeneous nature of the diseases, even if patients having similar
pathological features are treated similarly, the disease prognosis
differs a lot. It shows the inadequacy of symptom-based diagnosis and
demands patient-specific analysis of omics data. Collective analysis of
these multi-dimensional omics data is referred to as “pan-omics”
([32]Sandhu et al., 2018) which are also considered as “big” data in
the context of biological data analysis. Pan-omics data enable us to
predict novel functional interactions between molecular mediators at
multiple levels. Also, these data have the potential to uncover crucial
biological observations into hallmarks and pathways that would
otherwise not be obvious through single-omics studies. Patient-specific
pan-omics data analysis is going to disclose the genetic, epigenetic,
and other functional profiles responsible for the disease of an
individual which might eventually lead to the development of
individualistic “precision medicine” and will provide right treatment
to right patient at right time.
Cancer is a leading cause of death worldwide, being responsible for 9.6
million deaths in 2018 ([33]Bray et al., 2018). Cancer is a
heterogeneous disease caused by aberrations of genes and proteins.
“Precision oncology” promises identification of disease subtypes,
specific biomarkers and subsequently prediction and translation toward
the development of treatment procedures. Pan-omics or multi-omics
analysis in breast cancer has revealed significant differences in
molecular subtype distribution ([34]Kan et al., 2018). Genomics and
transcriptomics analysis of breast cancer data of Korean and Caucasian
cohorts showed underlying molecular differences, which are responsible
for the occurrence of breast cancer at the younger age in the Asian
population compared to the western population ([35]Kan et al., 2018).
Multi-omics analysis extended to different types of cancers confirms
the existence of broadly two types of cancers, cancers caused by
recurrent mutations and cancers caused by copy-number variations
([36]Mcgrail et al., 2018). Computational methodologies like,
artificial intelligence are being used widely to extract
patient-specific information from these big data, discussed in a recent
review ([37]Biswas and Chakrabarti, 2020). Machine learning based
pan-omics analysis of pan-cancer data shows the existence of clusters
within different types of cancers ([38]Ramazzotti et al., 2018),
identifies cell-model selective anti-cancer drug targets for breast
cancer ([39]Gautam et al., 2019).
Multiple data portals like TCGA ([40]TCGA, 2020) and ICGA ([41]Zhang et
al., 2011) have been developed to make multi-omics data conveniently
accessible. LinkedOmics contains pan-omics data of several types of
cancers ([42]Vasaikar et al., 2018). Databases like, GliomaDB ([43]Yang
et al., 2019) and MOBCdb ([44]Xie et al., 2018) are dedicated to
integrate multi-omics data for specific type of cancers. Standalone
software packages and web-servers are also being developed for the
analysis pan-omics data. [45]Table 1 compares the analytical tools
which are being used by researchers. R package mixOmics ([46]Rohart et
al., 2017), based on multi-variate analysis is available for the
integration of multi-omics data. It finds subsets of important features
but excludes network analysis. OmicsNet provides a web-based platform
to create different types of interactive molecular interaction networks
for single or multiple types of omics data ([47]Zhou and Xia, 2018).
Network-based integration of multi-omics data using iOmicsPASS, allows
to predict subnetworks of molecular interactions within a single type
or multiple types of omics data ([48]Koh et al., 2019). R package
Miodin ([49]Ulfenborg, 2019) provides a software infrastructure for
vertical and horizontal integration of multi-omics data but lacks a
comprehensive network analysis and visualization. PaintOmics allows
integrated visualization of multiple types of omics data in KEGG
pathway diagrams ([50]Hern et al., 2018). Software package, Multi-Omics
Factor Analysis (MOFA) ([51]Argelaguet et al., 2018) integrates omics
data in an unsupervised approach implementing generalized principal
component analysis (PCA). pathfindR ([52]Ulgen et al., 2019) finds
active sub networks for genes in omics data and perform pathway
enrichment analysis. R package Mergeomics ([53]Shu et al., 2016)
provides a pipeline to identify important pathways and key drivers in
biological systems. However, platforms required for systematic analysis
of the landscape of genetic, epigenetic, and metabolomics alterations
and biological and clinical relevance of multi-layer signature in
cancers are still limited.
TABLE 1.
Comparison of APODHIN with other existing pan-omics data analysis
tools.
Feature APODHIN OmicsNet [14] mixOmics [13] iOmicsPASS [15] Miodin [16]
Platform Web Web Standalone Standalone Standalone
Programming language Python, R, perl R R C^++ R
Types of omics data as input
mRNA transcriptomics Yes Yes Yes Yes Yes
miRNA transcriptomics Yes Yes Yes No Yes
Proteomics Yes Yes Yes Yes Yes
Phospho-proteomics Yes No Yes No No
Genomics Yes Yes Yes Yes Yes
Epi-genomics Yes No No No Yes
Metabolomics Yes Yes Yes No No
Multiple lists of same type of omics data Yes No Yes Yes Yes
Finding deregulated proteins/genes/miRNAs Yes No No No No
Map in meta-interactome Yes No No No No
3D interactive network Yes Yes No No No
Network topology analysis Yes No No No No
Prognostic status of proteins/genes (in cancer) Yes No No No No
Pathway enrichment analysis Yes Yes Yes Yes No
Analysis for regulatory network protein links Yes No No No No
[54]Open in a new tab
Different types of omics data carry information on different types of
bio-entities, e.g., genes, proteins, miRNAs, metabolites, etc. Hence,
integrative analysis of pan-omics data needs a meta-interactome
consisting of a protein–protein interaction network (PPIN) as well as
different regulatory networks. The web server for the Analysis of
Pan-omics Data in Human Interactome Network (APODHIN) provides a unique
platform where users can analyze different types of omics data using a
human cellular meta-interactome network. Graph theory based network
analysis has become an essential tool for analysis of PPIN for
extracting proteins important in the construction and information flow
of the network ([55]Jeong et al., 2001; [56]Barabási and Oltvai, 2004;
[57]Mistry et al., 2017; [58]Ashtiani et al., 2018), APODHIN provides
options to identify topologically important nodes (TINs) such as hubs,
bottlenecks, and central nodes (CNs) and their subsequent modules via
protein–protein interaction (PPI) and regulatory relationship network
analyses and pathway enrichment analysis. TINs are also correlated as
prospective diagnostic and/or prognostic biomarkers. APODHIN can also
analyze and compare multiple omics data set for a single omics layer,
such as transcriptomics, proteomics data collected from different
patient cohorts and/or different stage/grade of the same cohort.
Additionally, utilizing multi-omics data APODHIN calculates
cross-pathway regulatory and PPI links connecting signaling proteins or
transcription factors (TFs) or miRNAs to metabolic enzymes and their
metabolites using network analysis and mathematical modeling. These
cross-pathway links were shown to play important roles in metabolic
reprogramming in cancer scenarios such as glioblastoma multiforme in a
previous work ([59]Bag et al., 2019).
In addition to the server part, APODHIN shares analysis of multi-omics
data from various cancer cell lines where TINs and cross-pathway links
were identified using publicly available omics datasets collected for
various gynecological cancers. APODHIN platform is freely available at
[60]http://www.hpppi.iicb.res.in/APODHIN/home.html.
Materials and Methods
Server Description
Analysis of Pan-omics Data in Human Interactome Network web server is
dedicated for the integration and subsequent analysis using single or
multiple types of omics data. For single type of omics data, APODHIN
can analyze multiple datasets (up to 3) which may correspond to either
different stages of a disease from a single cohort or from dataset
collected from multiple patient cohorts and/or cell lines.
For multiple types of omics data, APODHIN allows single input data file
for each type of omics data. Following sections briefly describe the
various analytical part of the APODHIN server.
Data Collection
Analysis of Pan-omics Data in Human Interactome Network web server is
preloaded with a human cellular meta-interactome network. This
meta-interactome consists of human protein–protein interaction network
(HPPIN), network of human miRNAs and their target genes and network of
human TFs and their target genes. The PPI data was collected from
STRING ([61]Szklarczyk et al., 2019) database (version 11).
Interactions having a medium threshold of experimental score ≥700 were
considered ([62]Ferretti and Cortelezzi, 2011) for construction of the
PPIN. Target gene information of miRNAs was collected from the TarBase
([63]Vergoulis et al., 2012) and miRTarBase ([64]Chou et al., 2016)
databases. From the TarBase database (version 6) we have taken reliable
interactions supported only by low-throughput experiments (e.g.,
reporter gene assay, western blot, qPCR, etc.) whereas miRNA target
interactions with strong confidence (i.e., validated by either of
report assay, western blot, qPCR experiments) from miRTarBase (version
6) were considered for APODHIN meta-interactome network. We trusted on
the more reliable low-throughput experimental data to build the parent
miRNA-target mRNA interactome network. We found 2492 target genes for
544 miRNAs creating 6917 interactions. TFs and their target genes were
downloaded from Human Transcriptional Regulation Interactions database
(HTRIdb) ([65]Bovolenta et al., 2012). We found 11887 target genes for
284 TFs creating 18153 interactions. These three networks were merged
together to form the APODHIN meta-interactome consisting of two types
of biomolecular nodes i.e., proteins/genes and miRNAs along with three
types of interactions, i.e., protein–protein, miRNA-target gene, and
TF-target gene, respectively.
Additionally, we have also included a network of metabolites as
substrate and product with their corresponding metabolic enzymes in the
APODHIN server. For constructing this network, we downloaded metabolic
reactions from MetaNetX database ([66]Moretti et al., 2016) and
extracted the metabolites along with the corresponding metabolic
enzymes and further filtered those enzymes and metabolites which have
been listed in the Human Metabolome Database (HMDB) database
([67]Wishart et al., 2018).
Pan-omics Data Integration and Meta-Interaction Network Extraction
In APODHIN web server, user can upload single or multiple types of
omics data. The server accepts RNA transcriptomics, miRNA
transcriptomics, proteomics, phosphoproteomics, genomics, epigenomics,
and metabolomics data. The current version of the server accept only
processed format of the omics data where differential
expression/abundance of corresponding biomolecules are provided with
logFC for defining up and down regulation of genes/miRNAs/proteins and
threshold probability or p-value. For RNA transcriptomics, miRNA
transcriptomics and proteomics data user should select threshold values
of logFC for defining up and down regulation of genes/miRNAs/proteins
and corresponding adjusted p-value. Uploaded files should contain list
of genes/miRNAs/proteins along with logFC and p-values. Sample file
formats for different omics data are provided in the APODHIN help page.
For genomics, epigenomics, and phosphoproteomics data, genes that are
mutated and/or methylated and proteins, which are phosphorylated are
considered, respectively. APODHIN help page also provides guidelines to
process GEO ([68]Barrett et al., 2013) transcriptomics data for using
in APODHIN. Packages and tools for GEO series data are also enlisted in
the APODHIN “Help” page. For other types, of omics data like,
proteomics, genomics, metabolomics, useful links for data processing is
provided in the APODHIN help page and it will be made more enriched
gradually depending on the requirements from users.
Analysis of Pan-omics Data in Human Interactome Network web server
extracts the interactome networks from the parent meta-interactome for
the genes, mRNAs, miRNAs, proteins, and metabolites that are either
deregulated or altered according to the user supplied single or
multiple omics data. It creates a filtered meta-interactome network
comprising of deregulated or altered nodes and their 1st or 2nd level
(as chosen by user) interactors and/or regulators. For metabolomics
data, the web server finds out the proteins linked with metabolites and
constructs network. These single or multi omics data specific
meta-interactome networks are subsequently displayed in an interactive
three-dimensional (3D) network viewer within the APODHIN server. For
creating omics data mapped network, and subsequently network analysis,
APODHIN does not provide any special weight or scores to any type of
omics data.
For the module “pathway connectivity analysis,” RNA transcriptomics,
miRNA transcriptomics, and proteomics data were considered as primary
data and submission of at least one of them is mandatory to define
deregulated miRNAs and/or genes/proteins. In case of “pathway
connectivity analysis,” the logFC values for each of the uploaded omics
data is normalized in the scale of −1 to +1 following Eq. 1,
[MATH:
logFC
normalized=logFC|logFC|max :MATH]
(1)
where positive and negative values indicate up and down regulated
entities, respectively. If more than one primary omics data, for
example, transcriptomics and proteomics are provided, APODHIN web
server sums up the normalized logFC values from the different omics
data for the same node (RNA/protein) and if the sum is non-zero,
gene/protein is considered deregulated. Primary omics data determines
whether the gene is deregulated or not. Also, if a gene is found not
altered in supplied primary omics data, APODHIN does not consider this
gene for further analysis, irrespective of its status in the supplied
secondary omics data. Details of the utilization of the normalized
omics values in mathematical modeling based pathway connectivity link
identification are provided later. In this module, the information on
metabolites for any enzyme can be obtained in the associated table on
selection of enzyme.
Network Analysis and Identification of TINs
Once the context specific meta-interactome network is formed via
utilization of user supplied single or multiple omics data, APODHIN web
server primarily finds three types of TINs, namely, hubs, CNs
([69]Bhattacharyya and Chakrabarti, 2015) and bottlenecks (BNs) ([70]Yu
et al., 2007). To find the important nodes, network and node indices
like degree, betweenness, closeness and clustering coefficients are
calculated from the extracted meta-interactome network. These node
parameters were calculated using previously reported methods and
protocols ([71]Bhattacharyya and Chakrabarti, 2015). For
transcriptomics and proteomics data, TINs are identified from the
expressed nodes only. For phosphoproteomics, genomics, epigenomics and
metabolomics data, TINs are identified from phosphorylated, mutated,
methylated proteins/genes and metabolic enzymes, respectively.
Hubs are nodes that have high degrees. Degree distribution is
normalized following Eq. 2,
[MATH:
xi,normalized=xixmaximumn :MATH]
(2)
where x[i] is degree value of a node i and x[maximum] is the maximum
degree of the network. APODHIN web server converts normalized degree
distribution to corresponding z-score distribution. The plot of
probability distribution function (PDF) of z-scores for all nodes in
network is sent to the user by email. This email shares intermediate
results only. From the plot of PDF, users are asked to provide the
threshold value for hub identification. After receiving the threshold
value, APODHIN initiates hub identification program. Nodes having
degree greater than the threshold value are considered as hub. It is
also mentioned in the help page. Scores concerning individual
centrality parameters like, betweenness, closeness and clustering
coefficients are calculated and the cumulative centrality scores (CCS)
are estimated by summing over the combined scores for first layer
interactors ([72]Bhattacharyya and Chakrabarti, 2015). CCSs are
normalized following Eq. 2 where x is equal to CCS. Normalized CCS are
converted into z-scores. The PDF of z-scores for all nodes of network
are sent to the user by email and CNs are chosen based on the user
provided threshold value of z-score following similar procedure as
mentioned while identifying hubs. Bottleneck nodes are characterized
based on their betweenness values. Normalized betweenness values were
obtained from Eq. 2 where x is betweenness and subsequently, converted
into z-scores. Similar to hubs, bottleneck nodes are also chosen based
on the user provided threshold z-score, chosen from the PDF plot of
z-score for all nodes.
Further, sub-network consisting of TINs and their first or second layer
interactors are constructed and displayed in an interactive
three-dimensional (3D) network viewer.
The overlap of TINs, as well as all nodes of the network, as prognostic
cancer marker is checked after extraction of prognostic marker
information from the Human Protein Atlas database (version 19)
([73]Uhlen et al., 2017). The prognostic data was obtained from
Kaplan-Meier survival analysis. The cancer type, for which prognostic
status have minimum p-value, is shown in the “Node information” table
in the page of “network view of identified important nodes.” On mouse
hover on the cancer type, more detail information for other cancer
types, is available.
Pathway Mapping and Network of Mapped Pathways
For each identified TIN, particularly for genes and proteins, APODHIN
maps the corresponding pathways listed in the KEGG database
([74]Kanehisa et al., 2017). APODHIN performs a hypergeometric Fishers
Exact test and selects enriched pathways satisfying p-value (p[HGD])
≤0.05 using the following contingency table and formula.
[MATH: [a b-ac d-c] :MATH]
Where,
a = Number of genes in the pathway.
b = Number of genes in the gene list.
c = Total number of genes in the pathway.
d = Total number of genes in all pathways in KEGG.
[MATH:
pHGD=(ba)(dc)(b+da+c) :MATH]
(3)
Further, a network representation of important nodes along with their
enriched mapped pathways is displayed in an interactive
three-dimensional (3D) network viewer. [75]Figure 1A shows the flow
chart of “pan-omics data mapping and network analysis” module of
APODHIN.
FIGURE 1.
[76]FIGURE 1
[77]Open in a new tab
Flow charts showing work flow in APODHIN web-server for module (A) data
mapping and network analysis and (B) pathway connectivity analysis.
Pathway Connectivity Analysis and Cross-Pathway Links
This module of the APODHIN web server aims to construct regulatory
interaction networks and subsequently identifies cross-pathway
interaction links connecting different cellular pathway proteins [e.g.,
signaling proteins (S)], regulatory proteins [e.g., transcription
factor (TF)] or miRNAs with metabolic pathway proteins (M).
For this purpose, APODHIN web server was preloaded with cross-pathway
links or paths where protein–protein interactors (P) connect X nodes (X
can be S or target gene of TF or target genes of miRNAs) with M
(metabolic) proteins. We have limited the number (n) of protein–protein
interactors (P) to a maximum value of three between X and M proteins.
This limit provides four types of paths, XM (n = 0), XPM (n = 1), XPPM
(n = 2), XPPPM (n = 3). These cross-pathway linking paths are filtered
and selected based on expression and/or abundance status of the
biomolecules supplied by user uploaded pan-omics data for a given
disease or context. The filtering criteria for any given path is set
when the terminal nodes are found to be deregulated and the remaining
nodes are at least expressed within the user provided single or
multi-omics datasets.
We implemented an established probabilistic approach based on the
Hidden Markov Model (HMM) ([78]Tuncbag et al., 2013; [79]Vinayagam et
al., 2014; [80]Bag et al., 2019) utilizing the information of
experimentally established PPIs and gene regulatory information to
extract novel paths and interconnections between regulatory nodes such
as signaling proteins, TFs and miRNAs and metabolic pathway proteins
(M). Within these important X-M pairs, important cross-pathway
connecting paths are again scored by considering all filtered paths
between X-M pairs. To find important X-M pairs, weights are assigned on
nodes and edges depending on network and biological properties. Edge
weight is assigned in terms of normalized interaction probability which
is proportional to the product of their expression scores.
Two types of node weights, network entropy, and effect-on-nodes are
considered. Network entropy includes local entropy of the node. Another
node weight parameter, “effect-on-node” considers the impact of
interactors of a particular gene in the cross-connected network. The
“effect-on-node” considers both biological and network properties of
the node. Biological properties include deregulated gene, signaling
crosstalk gene and rate limiting enzyme. Network properties include
hubs, CNs, and bottlenecks.
Analysis of Pan-omics Data in Human Interactome Network web server
allows the user to choose maximum four weight options out of the six
weights. If a node satisfies any of the selected weight options, weight
value 1 is assigned for each satisfied option. To identify important
cross-connecting X-M pairs we have evaluated “path score” (PS) based on
a HMM implemented within the core mathematical model that calculated
the significant cross-pathway linking paths. “Path scores” are
converted to z-scores and paths having z-score ≥1 are considered as
important cross-connecting paths. A detailed description of the
mathematical models and path calculation is available in our previous
publication ([81]Bag et al., 2019). [82]Figure 1B shows the flow chart
of “pathway connectivity analysis” module of APODHIN.
APODHIN Architecture
Analysis of Pan-omics Data in Human Interactome Network web server is
created using HTML, PHP, PYTHON, and JAVA scripts. Client/user side
scripts are written in HTML, PHP and JAVA scripts. User uploaded data
is analyzed using PYTHON scripts. For network analysis, PYTHON package
networkX (version 1.8.1) is used. For visualization of 3D presentation
of networks JAVA scripts based open source technologies (three.js and
3d-force-graph.js) were utilized.
Analysis of Pan-omics Data in Human Interactome Network has two
separate parts A. APODHIN server and B. APODHIN example data analysis.
APODHIN Server
Analysis of Pan-omics Data in Human Interactome Network web server is
preloaded with human interactome network containing PPIN, target gene
network of miRNAs and target gene network of TFs. Proteins
participating in signaling and metabolic pathways are also marked
separately. Metabolites along with their target enzymes are also
included within APODHIN. This meta-interactome network is used as
framework of cellular interactions and is further used to map user
supplied single or multiple types of “omics” data to perform the
following analyses.
* •
Omics data mapping and network analysis: This module has two
sub-modules. On clicking first submit button, this web server
provides meta-interactome network filtered by uploaded omics data
where deregulated and/or altered nodes along with their interactors
are included. Users can further proceed for finding important
interacting nodes from the “pan-omics” data mapped interaction
network by clicking second submit button. Tabular, graphical and 3D
network representations of the identified TINs are provided for
better appreciation. Overlap of the TINs is shown both in tabular
and interactive 3D network visualization. Additionally, TINs and
their enriched pathways are also shown in tabular and interactive
3D network visualization manner.
Sample input files for each omics data type and example analysis
output are provided for the ease of use and apprehension.
* •
Pathway connectivity analysis: As mentioned before, this sub-module
highlights significant PPI and regulatory paths connecting
signaling proteins/TF/miRNAs to metabolic proteins. These
cross-pathway links are thought to be supra-molecular regulatory
links/signatures connected with metabolic rearrangement or
reprogramming events that are observed during cancer. In APODHIN,
these cross-pathway regulatory links can be constructed from three
types of interaction networks.
* 1.
Integrated network where signaling (S) and metabolic (M) pathway
proteins are connected through protein–protein interactors (P).
* 2.
Integrated network where target genes of TFs and metabolic (M)
pathway proteins are connected through protein–protein interactors
(P).
* 3.
Integrated network where miRNA target genes and metabolic (M)
pathway proteins are connected through protein–protein interactors
(P).
Cross-pathway linking paths are filtered and selected based on
expression and/or abundance status of the biomolecules supplied by user
uploaded single or pan-omics data for a given disease or context. These
paths are shown both in tabular and interactive 3D network
visualization.
APODHIN Example Data Analysis
Analysis of Pan-omics Data in Human Interactome Network example data
analysis page showcase few example analysis of multi-omics data for
different cancer cell lines. We have used the APODHIN web server to
construct individual cancer and dataset centric meta-interactome
network using cell line specific single and/or multi-omics data
collected from various resources such as GEO ([83]Barrett et al.,
2013), PRIDE ([84]Perez-Riverol et al., 2019), publication reports and
data sources for cervical, ovarian, and breast cancers, respectively.
Further, these cancer and dataset specific meta-interactome networks
were analyzed and important interacting nodes and cross-pathway links
were identified and provided within the APODHIN example data analysis
module. We have used cancer cell line derived omics data freely
available from different public resources. Options are provided for the
users to select single and/or multi-omics data to construct the
meta-interactome networks and further analyze them to identify and
important interacting nodes and cross-pathway links specific for the
selected dataset.
Results
Input Options
Analysis of Pan-omics Data in Human Interactome Network server provides
two different but linked analysis options for the users who would like
to utilize single or multiple types of omics data for a given context.
APODHIN web server provides options to upload seven types of “omics”
data comprising of mRNA transcriptomics, miRNA transcriptomics,
proteomics, phosphoproteomics, genomics, epigenomics, and metabolomics.
The file formats for each data type is specified in the “Help” page and
sample input files are also available in the server input page.
Information on preparing input files for using in APODHIN is also
shared in the “Help” page. For transcriptomics and proteomics data,
maximum and minimum threshold values for the differential
expression/abundance (logFC) and statistical significance of that
(p-values) need to be provided. As the calculations are computation
intensive, results are sent via email.
Similarly, for cross-pathway connectivity analysis users need to upload
single or multiple types of “omics” data for a given context. At least
one “primary” type (see Methods) of omics data need to be uploaded.
Now, in this case, users also need to specify the type of connectivity
they would like to explore, for example, signaling to metabolic
proteins, TFs to metabolic proteins, or miRNAs to metabolic proteins.
Only one type of pathway connectivity can be explored at a time for a
given set of “omics” data. Additionally, users also need to select the
kind of weights (see section “Materials and Methods”) that would be
applied while calculating the scores of the selected cross-pathway
regulatory and PPI paths. E-mail address needs to be supplied for
APODHIN server to send the result link of the identified cross-pathway
connections.
Output Options
Output option for the “Data mapping and network analysis” module has
two stages. At first stage ([85]Figure 2A), the context specific
meta-interactome network (“filtered network”) can be visualized via a
user interactive 3D network viewer where information regarding each
node and edge are provided in graphical as well as tabular view
([86]Figure 2B). Status of the “omics” data mapping is shown in various
color codes for the nodes whereas different relationship like PPI,
miRNA-target gene interaction, and TF and target in connections are
shown varied color codes. Additional details about the protein nodes
can be obtained via GeneCards ([87]Stelzer et al., 2016) link while
miRNA details can be found via miRTarBase ([88]Chou et al., 2016) link.
List of metabolites mapped onto the protein nodes are also provided
both in the network viewer as well as in the adjacent tabular format.
If network analysis is opted, along with filtered network, APODHIN
provides the PDFs for the opted TINs ([89]Figure 2C). Filtered nodes
(genes/proteins/miRNAs) that satisfied the selected threshold criteria
are characterized as TINs and further utilized for meta-interactome
network construction. If multiple files of single type of omics data is
uploaded, users can see the number of TINs (as hub, bottlenecks, and
CNs) and their mutual overlap using interactive Venn diagram by
clicking the “link for analysis” option for single or combination of
“omics” data ([90]Figure 2D). Combined analysis of multiple types of
omics data files is shown if multiple types of omics data files are
provided. Here also, the resultant page ([91]Figure 2E) provides three
output options. First, the regulatory and PPI connectivity specific to
the hubs, bottleneck and CNs can be seen via corresponding link where
networks of deregulated hubs, bottleneck, and CNs can be seen
separately and saved accordingly ([92]Figure 2F). Association to
various kinds of cancers for the identified TINs as
favorable/unfavorable prognostic markers are also provided here after
mapping the TINs (see Methods) to the data provided in Human Protein
Atlas ([93]Uhlen et al., 2017). Another option provides the network of
common TINs ([94]Figure 2G) whereas a separate link provides network of
enriched pathways with the identified TINs ([95]Figure 2H). Enriched
pathway networks of deregulated hubs, bottleneck, and CNs can be seen
separately and saved accordingly. In all these three network output
options, data can be downloaded in text format for further analysis.
FIGURE 2.
[96]FIGURE 2
[97]Open in a new tab
Snapshots of outputs of module “data mapping and network analysis.” (A)
Page showing link for filtered network and probability distribution
function. (B) Filtered network. (C) Probability distribution function
for network analysis. (D) Output page of a single omics data. (E)
Network analysis page for multi-omics data. (F) Network of important
interacting nodes. (G) Network of important nodes. (H) Network of
pathway mapping.
Similar to “Data mapping and network analysis,” “Pathway connectivity
analysis” module also provides a tabular result with a summary of the
user uploaded data ([98]Figure 3A). Users can see the cross-pathway
links for multiple types of omics data ([99]Figure 3B). For multiple
types of files with single type of omics data ([100]Figure 3C), the
comparison ([101]Figure 3D) is shown in Venn diagram as well as in
network visualization. In the 3D network visualization window,
significant PPI and regulatory paths connecting signaling
proteins/TFs/miRNAs to metabolic proteins are shown in color coded
fashion. As described before, these cross-pathway links or paths
connect X nodes (X can be S or target gene of TF or target genes of
miRNAs) with metabolic (M) proteins. These linking paths are filtered
and selected based on expression and/or abundance status of the
biomolecules supplied by the users where for any given path the
terminal nodes are found to be deregulated and the remaining nodes are
at least expressed. The corresponding pathways and biological functions
of the proteins are also provided in tabular format adjacent to the
network viewer. Additionally, the metabolites connected to the
metabolic proteins that are part of the selected cross-pathway links
are also provided in the same page.
FIGURE 3.
[102]FIGURE 3
[103]Open in a new tab
Snapshots of outputs of module “pathway connectivity analysis.” (A)
Output page shows user provided data in tabular form along with link
for network view. (B) Output page showing network of signaling to
metabolic proteins connecting paths for multiple types of omics data.
(C) Output page when multiple files for single type of omics data is
provided. (D) Venn diagram shows overlap of signaling to metabolic
proteins connecting paths for different omics data set.
Example Data Analysis Option
Analysis of Pan-omics Data in Human Interactome Network example data
analysis page contains important nodes (genes/proteins/miRNAs),
pathways, and their networks with interacting partners specific for
cancers affecting women such as cervical, ovarian, and breast cancer.
This section also contains important paths linking signaling
proteins/TFs/miRNAs to metabolic enzymes, which could perhaps be
responsible for metabolic reprogramming in cancer. The example content
is produced by APODHIN web server using publicly available cervical,
ovarian, and breast cancer specific cell line based omics data.
[104]Figure 4 briefs the statistics derived from APODHIN example
analyses for mRNA transcriptomics data of different cell lines of
cervical, ovarian, and breast cancer. [105]Figure 4A shows the overlap
of deregulated genes. It reveals lesser overlap among deregulated genes
across cell lines for all cancers. However, there is almost complete
overlap of pathways mapped by deregulated genes ([106]Figure 4B). Nodes
satisfying any two types of TINs are considered as important
interacting nodes (IINs). [107]Figure 4C shows overlap for common IINs
between cell lines across cancer types are observed. Similarly,
[108]Figure 4D shows much higher overlap of common pathways mapped by
IINs. This demonstrates that IINs and their pathways represent the
common core genes and processes related to a cancer type in a better
way than that achieved by the initial deregulated genes obtained from
the omics data. We also checked whether the mapped pathways are related
to cancer pathways enlisted in KEGG database ([109]Kanehisa et al.,
2017). [110]Figure 5A shows that pathways mapped by IINs are more
cancer specific compared to the pathways mapped by deregulated genes
for all cell lines. [111]Figures 4E,F show the number and overlap of
deregulated genes and IINs as prognostic markers of respective cancer
type. [112]Figure 5B shows that compared to the deregulated genes, IINs
possess higher fractions of prognostic markers for all cancer cell
lines, except MDAMB231. This advocates the usefulness of the IINs over
deregulated genes. Moreover, as the number of IINs is much smaller than
that of deregulated genes the false discovery rate is also expected to
be lower.
FIGURE 4.
[113]FIGURE 4
[114]Open in a new tab
Statistics derived from APODHIN database for mRNA transcriptomics data
derived from different cell lines of cervical (HeLa, SiHa, and CaSki),
ovarian (IGROV1, SKOV3, OVCAR3), and breast cancer (MCF7 and MDAMB231).
Transcriptomics data was derived from the GEO datasets [115]GSE9750,
[116]GSE19352, and [117]GSE71363, respectively. (A) Deregulated genes,
(B) Overlap of pathways mapped by deregulated genes, (C) Overlap of
IINs, (D) Overlap of pathways mapped by IINs, (E) Deregulated genes as
prognostic marker, and (F) IINs as prognostic marker.
FIGURE 5.
[118]FIGURE 5
[119]Open in a new tab
(A) Comparison of number of cancer specific pathways mapped by
deregulated genes and IINs. (B) Comparison of fraction of prognostic
markers within the deregulated genes and network analysis derived
important nodes, such as IIN and various TIN (e.g., Hubs, CN, and BN,
respectively). Dashed lines are drawn to separate cell lines of
different cancer types.
[120]Figure 6 shows overlap of cross-pathway links or paths connecting
signaling (S) proteins, TFs, and miRNAs to metabolic (M) proteins
identified using omics data derived from the cell lines of three types
of cancers. For signaling to metabolic connection, four common paths
for three cervical cell lines were observed. However, no such overlap
was found for breast and ovarian cancer cell lines.
FIGURE 6.
[121]FIGURE 6
[122]Open in a new tab
Statistics derived from “pathway connectivity analysis” module of
APODHIN database for mRNA transcriptomics data derived from different
cell lines of cervical (HeLa, SiHa, and CaSki), ovarian (IGROV1, SKOV3,
OVCAR3), and breast cancer (MCF7 and MDAMB231). Transcriptomics data
was derived from the GEO datasets [123]GSE9750, [124]GSE19352, and
[125]GSE71363, respectively. (A) Overlap of cross-pathway links
connecting signalling (S) to metabolic (M) proteins, (B) Overlap of
pathway links connecting target genes of TFs to metabolic (M) proteins,
and (C) Overlap of cross-pathway links connecting target genes of
miRNAs to metabolic (M) proteins.
Analysis of pan-omics data considering transcriptomics, genomics,
epigenomics, metabolomics data in different combinations are available
for different cell lines in the example data analysis section of
APODHIN.
Discussion
Large-scale genomics, transcriptomics and proteomics approaches have
made it possible to characterize different clinical spectra associated
with cancers. Use of pan-omics platforms and approaches in the analysis
of systemic disease like cancer will not only help to identify numerous
useful biomarkers but also will expose areas for further improvement in
therapeutic intervention. Here, we present APODHIN web server, which
extracts cellular interactome networks from the parent meta-interactome
for the genes, mRNAs, miRNA, proteins, and metabolites that are either
deregulated or altered according to the user supplied single or
multiple omics data. These single or multi-omics data specific
meta-interactome networks are utilized to identify TINs and their
sub-modules enriched with PPI and regulatory relationship via
utilization of graph theory based network analyses and biological
pathway enrichment analysis. Important interacting nodes (proteins and
miRNAs), IINs are identified based on the overlap of key nodes such as
hubs and bottlenecks. Using data from The Human Protein Atlas database,
APODHIN provides the probable prognostic status of the IINs. Also, as
observed in our earlier works ([126]Bhattacharyya and Chakrabarti,
2015), IINs extracted from network topology, could correlate to be
prospective diagnostic and/or prognostic biomarkers or even turn out to
be potential therapeutic targets.
Molecular mechanisms for cancer progression and development of
potential therapeutics to inhibit these complex diseases are difficult
from the independent knowledge of signaling, TFs, miRNAs, and metabolic
pathways. Metabolic reprogramming is an essential hallmark of cancer
([127]Hanahan and Weinberg, 2011). Understanding the coordination among
various cellular pathways, such as gene-regulatory, signaling and
metabolic pathways is crucial and may provide clues into the molecular
mechanism of metabolic adaptation in cancer and associated cells.
Therefore, there is an urgent need for systems biology model, which can
coordinate among signaling-induced proliferation of tumor cells/growth,
transcription factor/miRNA based gene regulation and metabolic
processes. Hence, we emphasized to design a mathematical approach to
identify significant proteins forming interconnections between
signaling, regulatory and metabolic pathways. We have constructed an
integrated network where signaling (S), regulatory (TFs and miRNAs),
and metabolic (M) pathway entities are connected through
protein–protein and gene regulatory interactions. Interconnections
between regulatory components such as signaling proteins/TFs/miRNAs and
metabolic pathways need to be elucidated rigorously to understand the
role of oncogene and tumor suppressors in regulation of metabolism
alongside their normal functions. Analyses of such cross-connected
network and linking paths will facilitate probable way(s) to inhibit
cancer progression in a more specific manner.
Considering the growing demand of multi-omics data integration followed
by systems biology based analytical interpretation of the large-scale
“omics” data, implementation of a robust and user-friendly web-based
platform is very much due. In order to make better sense out of the
various “omics” data, it is imperative to utilize them in a way so that
the global scenario of the complex and multi-layer cellular interactome
can be recapitulated. Several data portals have been coming up to make
multi-omics data accessible, visible and more importantly,
interpretable. Various programs and web portals are being made to
interpret omics data in different perspectives. Each of these tools has
its own merits and limitations also. [128]Table 1 provides a
qualitative comparison of features and functionalities of APODHIN with
respect to existing omics data analysis tools. Web servers like
OmicsNet ([129]Zhou and Xia, 2018) is a technically powerful web based
platform specifically meant for better visualization of molecular
networks. It mainly provides varied and efficient ways of network
visualization including different components. However, it provides
minimal emphasis on networks analysis and identification and
interpretation of important interacting nodes and cross-pathway links.
Similarly, this server only accept differential omics data for
genes/proteins and metabolites, it does not have the option to include
the epigenetic modification, miRNA expression data, and
phosphoproteomics data. mixOmics ([130]Rohart et al., 2017) is a
software package which is based on multi-variate analysis. It performs
data reduction, and then identifies combination of biomarkers. It
offers a network visualization but does not consider network topology.
It does not consider any meta-interactome. Software package iOmicsPASS
([131]Koh et al., 2019) considers a meta-interactome by including PPIN
and TF regulatory network within omics data. But it excludes miRNA-mRNA
regulatory network. It considers only three types of omics data,
transcriptomics, proteomics, DNA copy number data, thus limiting its
applicability. Another R package Miodin ([132]Ulfenborg, 2019) provides
opportunity of creating a workflow of data analysis. It considers
different omics data, but not metabolomics data. It requires
pre-installation of several R packages. Miodin provides Venn diagram of
differentially expressed genes, overlapped within different datasets.
However, Miodin does not consider any meta-interactome and does not
construct any network. None of these tools perform network topology
analysis and provide cross-pathway connectivity information of
proteins. APODHIN is perhaps the only available web based platform that
offers to (a) integrate multi-omics data onto an exhaustive multi
layered cellular meta-interactome network, (b) extract and analyze the
context specific networks and sub-networks to identify TINs that could
serve as potential biomarkers and/or therapeutic targets (c)
rationalize the role of the identified TINs to the given context via
pathway enrichment and prognostic marker correlation, and (d) identify
cross-pathway interconnections between regulatory components such as
signaling proteins/TFs/miRNAs and metabolic pathways for better
understanding the role of oncogenes and tumor suppressors in regulation
of metabolic reprogramming during cancer. Additionally, being a web
based tool, APODHIN requires no installation of software, good
computing systems, and technical expertise. We believe these features
make APODHIN useful as well as a user-friendly application.
However, there is still scope for improvement for the APODHIN server.
The example data analysis part can be enriched to upgrade as a
database. For example, in future we would like to equip the server to
accept and process raw “omics” data directly and further create the
processed data for genetic or epigenetic alterations, differential
expression and abundance, respectively. We would also like to add
components for handling large number of datasets which will be able to
analyze cohort data. Current version is mostly aimed to
patient-specific personalized data. Similarly, the server and along
with a database should be enriched in such a way that it could be
utilized for deep learning and artificial intelligence based tools to
predict the disease outcome, recurrence and drug resistance,
respectively.
Data Availability Statement
The datasets presented in this study can be found in online
repositories. The names of the repository/repositories and accession
number(s) can be found in the article.
Author Contributions
NB and SC designed the web server. NB created the web server. KK, SB,
and RB provided the data for meta-interactome network. NB and SC
analyzed the data and drafted the manuscript. All authors contributed
to the article and approved the submitted version.
Conflict of Interest
The authors declare that the research was conducted in the absence of
any commercial or financial relationships that could be construed as a
potential conflict of interest.
Footnotes
Funding. The authors acknowledge CSIR-Indian Institute of Chemical
Biology for infrastructural support. SC acknowledges the Systems
Medicine Cluster (SyMeC) grant (GAP357), Department of Biotechnology
(DBT) for funding. NB acknowledges the Systems Medicine Cluster (SyMeC)
grant (GAP357), Department of Biotechnology (DBT) for fellowship. KK,
SB, and RB acknowledge Department of Biotechnology (DBT), Council of
Scientific and Industrial Research (CSIR), respectively for their
fellowships. The funders had no role in study design, data collection
and analysis, decision to publish, or preparation of the manuscript.
This manuscript has been released as a pre-print at bioRxiv
([133]Biswas et al., 2020).
References