Page 13 - GPD-2-1

Gene & Protein in Disease AI-based drug repositioning

can be obtained within their own specified range . A new and missing data can occur among samples, there is a
[35]
method called DRIMC was proposed by Zhang et al. The deviation between the calculated and actual results of the
DRIMC method integrates data from multiple sources such matrix approach.
as drugs and diseases, while using inductive matrices to Hence, it is recommended that researchers combine
complete the modeling of relevant probabilities . Peng et different strategies and methods to achieve higher
[36]
al. proposed a drug-target relationship prediction method rates of success. The effective combination of different
based on deep forest and positive and unlabeled (PU) methodological strategies and available data will also
learning, which constructs a similarity matrix between lead to great advances in the field of drug repositioning.
drugs and a similarity matrix between targets based on drug As artificial intelligence technology develops, more and
structure information and target sequence information, more effective ways will emerge to help understand disease
respectively . Yi et al. developed a multiple attempt
[37]
learning based on matrix completion for drug repositioning mechanisms and develop appropriate treatments. More
method multiview learning with matrix completion algorithms being used in the drug development process
(MLMC) . They used multiple view learning so as to in the future, combined with the foundation of traditional
[38]
predict new indications, while using matrix completion biological experiments, will be the basis for newly
for the associated sparse matrices for preprocessing so developed drugs with greater relevance and adaptability to
that features between multiple similarity matrices can be the human body.
computed. First, to calculate the best similarity matrix, Acknowledgments
they used multiview learning to predict multiple disease
similarity matrices and multiple drug similarity matrices. None.
Second, to make the multiview learning predictions more
accurate, the values of the related matrices were populated Funding
using matrix complementation methods. Finally, the above This work is supported by the National Natural Science
two steps were merged into one strategy in MLMC. The Foundation of China (Grant no. 62072157, 61802116) and
execution flow of MLMC is shown in Figure 4. the Natural Science Foundation of Henan Province (Grant
no. 202300410102).
4. Conclusion
This paper presents the research progress of artificial Conflict of interest
intelligence-based drug repositioning, focusing on The authors declare that they have no competing interests.
network-based approach, feature-based approach, and
matrix-based approach. Author contributions
Each method of AI-based drug repositioning has its Conceptualization: Qingkai Hu and Xianfang Wang
advantages and disadvantages. Network-based approaches Visualization: Yifeng Liu, Yu Sang, and Dongfang Zhang
are simple and reliable and are able to explore disease-drug Writing – original draft: Qingkai Hu and Xianfang Wang
target network relationships, but they cannot predict the Writing – review & editing: Qingkai Hu and Xianfang Wang
targets of new drugs and are very limited. However, network-
based approaches have great potential for deciphering the Ethics approval and consent to participate
underlying mechanisms of complex diseases, the mode Not applicable.
of action of drugs, and for repositioning disease-specific
drugs. With feature-based approaches, drug development Consent for publication
takes relatively long because the data requirements are Not applicable.
relatively high and require specialized expertise to design
the label. In particular, the development of robust model Availability of data
for feature-based computational drug repositioning is
a very complex process. One of the biggest difficulties is Not applicable.
to put theoretical computational approach into practice, References
because mapping between the theoretical approach and
the behavior of biological organisms is more complex. 1. Peng C, Hu Y, Chen L, et al., 2020, A review of drug
As for matrix-based methods, because they do not rely repositioning algorithms based on machine learning and big
on feature extraction and negative sample selection, they data mining. Adv Pharm, 44(1): 6.
do not require setting labels and have a relatively short 2. Zhang W, Gu F, Fu YK, et al., 2021, Research progress of drug
development time. However, inaccuracy, extreme data, repositioning in new drug development. Anim Husbandry

Volume 2 Issue 1 (2023) 7 https://doi.org/10.36922/gpd.v1i3.201