artificial intelligence (AI)). These representations include interactions presented

between genes, proteins, targets, drugs, and diseases in various combinations

(Yella et al. 2018). Novel information can arise from such analysis as formerly

unstudied, yet potentially accurate interactions may result, hence increasing the

opportunity for drug candidates that were previously not considered.

There are many network-based methods that have been developed for pathway

analysis (Nguyen et al. 2018). These are commonly found to be topology-based

commercial tools such as MetaCore and iPathwayGuide. Individual databases that

can be used to perform in silico analysis for DR include disease-based databases

such as the Cancer Genome Atlas (TCGA) and the Cancer Cell Line Encyclopedia

(CCLE) (Kwon et al. 2019). These data sets compile and hold the gene expression

profiles and additional research information obtained from in vitro, in vivo, and

clinical samples. Some of the drug-based databases include CMap, LINCS, and

CTRP, which contain much information of drug features and efficiencies (Kwon

et al. 2019). Knowledge-based databases include the Gene Ontology, KEGG, and

MSigDB, which display all studied mechanisms or pathways available in literature

(Kwon et al. 2019). To bring these data sets closer to DR, wet-lab-based DR tools are

available, which allow the exploration of the aforementioned data sets altogether.

Despite this, such tools often require proficient computer skills and high-

performance computing resources (Kwon et al. 2019). Examples of such tools

include the CLUE, L1000CDS², and DeSigN. Another online tool of such nature

is Gene2Drug, which uses pathway annotations from multiple sources (CMap, GO

BP, GO MF, CP, KEGG, Biocarta, Reactome, CGP, TFT, CORUM) to identify one

or more candidate drugs that are able to modulate a therapeutic target for the disease

of interest (Napolitano et al. 2018).

DR using in silico methods are now improved with the aid of AI. The machine

learning algorithms that have been incorporated into some online tools have allowed

for enhanced analysis of large omics data sets (Koromina et al. 2019). Some well-

known tools and online platforms that are being used for DR include Biovista

(Biovista 2021) and drug repurposing (Koromina et al. 2019; Drug Repurposing

Online 2021). These AI-supported tools provide assistance in pinpointing

non-obvious correlations between targets of interest, thus predicting high potential

candidates for DR.

A recent study (Zeng et al. 2021) developed their own network-based deep-

learning approach named deepDR for the purpose of in silico DR. deepDR integrates

10 networks of drugs, diseases, and side effects to predict specific associations and

highly specific drugs. Several predictions using deepDR have already been validated

by the ClinicalTrials.gov database for some diseases such as Parkinson’s and AD

(Zeng et al. 2021). Other studies have integrated the deepDR as a basis to model their

DR usage. The deep2CoV framework model was developed to effectively search for

potential drugs for COVID-19 to reduce redundancies in clinical trials (Liu et al.

2020). This heterogeneous network was designed to predict candidate drugs using

drug-drug, drug-disease, and drug-target data sets. Using deep2CoV, the following

drugs were predicted as candidates for COVID-19: ceftriaxone, ciprofloxacin,

60

I. A. Farouk et al.