Chapter3 : Fighting COVID-19 with Artificial Intelligence

reading notes of《Artificial Intelligence in Drug Design》

---

1.Introduction

• Drug repurposing, also referred to as therapeutic switching, drug redirecting, repositioning, reprofiling, or re-tasking, is a strategy that focuses on the identification of novel uses and/or indications for established drug substances.

• This strategy offers serval advantages when compared to developing new molecular entities (NMs):

  • these drug have already passed toxicity tests
  • reduce time frame
  • lower cost of drug development

• Historically, the strategy of repurposing has been led more by chance than actual rational design, such as sildenafil and thalidomide.
• Drug repurposing is well suited for AI applications with respect to capturing informative features from noisy, incomplete, and high-dimensional data.
• The accelerated global spread of SARS-CoV-2 lead to an immense urgency for identification of new therapies. Drug repur- posing was considered a promising approach for two main reasons:
  • the similarity of SARS-CoV-2 with other coronaviruses and the relatively straightforward way of sample acquisition and study.
  • chemical and genomic data together with pharmacological and phenotypic information are rapidly accumulating and they are, for the most part, accessible and standardized.

2.Structure-Based Drug Repurposing

• One of the crucial identified targets for this disease is SARS-CoV-2 M^pro, which shares high sequence similarity with SARS-CoV and other proteases.

• Another promising drug target for COVID-19 is the trans- membrane serine protease family member II (TMPRSS2).

3.Application of AI in Drug Repurposing

• Drug repurposing is based on the premise that different diseases share one or more targets that are involved in a cascade of protein networks. This information can be retrieved from databases and literature and used by artificial intelligence algorithms to identify common pathways and, thus, their relative drugs on the market.

3.1.Deep Learning

• In another study, Random Forest (RF) classification models were used to identify infected cells from their morphological profile and experimental tests of 1425 FDA-approved drugs identified lactoferrin as promising SARS-COV-2 inhibitor. CNNs and RNNs have been combined to predict antiviral drugs that could be repurposed against COVID- 19.

3.2.Graph Representation Learning or Network Proximity Analysis

• This algorithm combines AI with network medicine, by building up knowledge-based medical graphs, which connect diseases, targets and drugs in relationship networks. This method can be applied to predict new connections between a novel disease and existing drugs.

4.Repurposed Drug Investigation

• A proportion of promising drugs were predicted by artificial intelligence protocols as potential COVID-19 treatments. Here we list few of them:
  • Antiviral: remdesivir, atazanavir, efavirenz, ritonavir, and dolutegravir
  • Inhibitors of mRNA translation: zotatifin
  • Rheumatoid arthritis: baricitinib
• Zhou et al. used a network-based approach to identify toremifene and emodin, sirolimus and dactinomycin, mercaptopurine, and melatonin as potential drug combinations, based on the analysis of the drug targets that are involved in the interaction between the virus and the host.

5.Challenges and Future Perspectives

• Drug repurposing is a relatively swift method for identifying promising antiviral treatments during a pandemic. Still one should consider that drugs on the market currently are optimized for a specific effect and medical use, such as dosage.

• Tracking of mutagenesis data across international patient records and predictions of coding sites with high functional relevance and risk of mutations can be performed by AI and enable the development of effective therapeutics and vaccines.

• Out of 4300 clinical studies, half of them are in the process of recruiting participants who need to match specific requirements. With the aim of making this process faster and more efficient, instead of a randomized selection, machine learning can be used for patient group stratification by selecting subgroups of patients for clinical trials

• Given the physiological complexity of individual patients, artificial intelligence offers a strategy to make personalized treatments possible.