Page 8 - GPD-2-1

Gene & Protein in Disease AI-based drug repositioning

The recent drug development case of COVID-19 Many network-based drug repositioning methods have
[4]
is a typical case of faster and further exploration of drug been proposed by researchers in recent years, as shown in
repositioning. Traditional new drug development requires Table 1.
a lot of investment, takes a long time, and is risky. With A network-based inference method was invented by
the help of artificial intelligence technology, virtual Cheng et al. . To derive new targets for already marketed
[7]
high-throughput screening of candidate compounds drugs, this method only needs to use the drug-target
can be performed, thus enhancing the efficiency of dichotomous network topological similarity. Guney et al.’s
drug development. The techniques related to artificial team used the metric of disease-drug similarity to calculate
intelligence are applied in different aspects of drug the magnitude of the interaction between a disease and
repositioning to solve many key problems . For example, a drug target . The method is highly systematic and
[5]
[8]
active compound screening, molecule generation, drug comprehensive by introducing chemical similarity for
target discovery, and protein structure and protein-ligand correlation and by considering the necessary biological
interaction prediction are widely used. information. Wang et al. team invented a heterogeneous
In this paper, we introduce the research progress of drug network modeling framework that computes by capturing
repositioning in recent years, focusing on three categories: various interrelationships between targets, drugs, and
Network-based methods, feature-based methods, and diseases with each other to predict the effectiveness of new
matrix-based methods, as shown in Figure 1. drug use .
[9]
2. Net-based approach On the basis of similarity-based heterogeneous networks,
deep learning techniques can be used to represent proteins
For modeling biological and biomedical entities, and their
relationships and interactions, networks are the best way to go. Table 1. Net‑based approach
Networks can provide models of how drugs and indications,
as well as drug targets, work to determine therapeutic drug Methods Features References
potential. When representing biological data by networks, NBI Using only drug-target dichotomous [7]
usually genes, molecules, proteins, and other biological network topological similarities to infer
entities can be represented by nodes; and their relationships new targets for known drugs
such as mode of action, similarity, association, and Drug-disease A drug-disease similarity metric was [8]
interaction are represented by edges . For specific attribution proximity introduced
[6]
information, it is generally represented by the weighted values TL_HGBI Using Disease Information to Predict New [9]
of edges and node; examples include gene-gene interaction Drug Targets
networks, networks of drug-target interactions, and networks NRWRH Enables large-scale prediction of potential [10]
of interactions between various other biological entities. For drug-target interactions
learning graphical data with nonlinear relationships, the DTI-CNN Drug-target interactions based on feature [12]
graphical neural network in neural networks can be used; the representation learning and deep neural
networks
network can also be used to represent biological entity data.
Moreover, a network-based chemical similarity correlation NBI: Network-based inference, TL_HGBI: Triple-layer heterogeneous
graph-based inference, NRWRH: Network-based random walk
analysis method can be used to discover side effects of new with restart on the heterogeneous network, DTI-CNN: Drug target
drugs as well as to reposition the already marketed drugs. interaction prediction.

Figure 1. Artificial intelligence technology applied in various aspects of drug repositioning.

Volume 2 Issue 1 (2023) 2 https://doi.org/10.36922/gpd.v1i3.201