Leveraging Machine Learning Applications for reducing the drug discovery time by Pragya Sharma – Student Author

The entire world has come to a standstill in 2020 due to the spread of COVID-19 pandemic. According to an article published in march 2020 the major cause of this panic is that COVID-19, the illness caused by a severe acute respiratory syndrome known as SARS-coronavirus 2, is much more  contagious and  deadly as compared to other infections, furthermore, there are no approved treatments or vaccines. Currently there are no FDA approved drugs or vaccines for the prevention and/or treatment of COVID-19 (1).

Drug discovery processes are tedious, expensive and it takes years for a new drug to get an approval. It’s well known that the success rate for drug development (as defined from phase I clinical trials till drug approvals) is very low across all therapeutic areas, across the global pharmaceutical industry. A recent study on 21,143 compounds found that the overall success rate was as low as 6.2%. The cost (for a drug company) to bring the drug to market in 2018 was approaching $2 billion for 12 years of development. Although more than $50 billion is spent on research and development (R&D) per year by large pharmaceutical companies alone, the FDA approves only 30 new chemical organizations per year.

With the advancement in the Artificial Intelligence (AI) and Machine Learning (ML), it should be possible to bring medicines to the market 500 days faster, which would create a competitive advantage within increasingly crowded asset classes, and bring much-needed therapies to patients, sooner. Drug development can be transformed by AI, which can reduce the development costs by 25 percent (2).

Structure of the molecules can be described as a molecular graph, which makes the use of graphical neural networks and other neural network-based techniques possible. In an article Attention and Edge Memory schemes were implemented to the existing message passing neural network framework, different physical–chemical and bioactivity datasets from the literature were used as standards (3). Another recently publishes article uses a graph-neural-network framework called self-attention-based message-passing neural network (SAMPN) to study the relationship between chemical properties and structures in an interpretable way (4).

ML and other tools are utilized to predict how the drugs will interact with the targets, and they can reduce the time taken by clinical trial experiments by 70 percent. Detection of drug activity and toxicity with a great precision as compared to other computational methods as well as a reduction in uncertainty by applying algorithms that repeatedly choose meaningful experiments based on emerging patterns is a competitive advantage of ML, for the pharmaceutical industry (2).

A.I. transforming clinical trials

There is no doubt that long-established clinical trials remain an acceptable way to ensure the effectiveness and safety of new drugs, they are deprived of the analytical power, flexibility and speed needed to develop new therapies aimed at overcrowding in younger and more frequent patients. Unlocking RWD (real-world data) using speculative AI models and analytics tools can accelerate disease comprehension, identify relevant patients and key investigators to inform site selection, and support the formation of a novel clinical survey. AI-enabled technology, has the ability to collect, organize and analyze a growing body of data generated by clinical trials, including failed, can produce sensible patterns of information that help to create clinical trials; AI-enabled digital transformation can improve patient choice and increase clinical trial efficiency, through mining, analysis and translation of multiple data sources (e.g. electronic health records). Other areas of clinical trials that can be transformed are investigator and site selection, patient monitoring, medication adherence and retention, using AI operational data enabled by clinical trial analysts, Outsourcing and strategic relationship acquisition of required AI skills and talent. In near future, the silico trials will soon be adopted using the advanced computer model and measurement in the development or legal testing of the drug (5).

AI- based Drug repurposing

Scientists and researchers have come up with a more practical and speedier alternative for drug discovery i.e. repurposing the already approved molecules by conducting in silico screening and classification of the drugs and the compounds that have the potential to denature the essential viral proteins (6).

Although the ultimate solution to the problem is developing an exclusive medicine or vaccine, repurposing of drugs is effective enough to minimize the time and costs in drug development since data is already available on their potential toxicity, formulation and pharmacology. Hydroxychloroquine and Remdesivir are some of the examples of repositioned drugs. In spite of the fact that the use of hydroxychloroquine is now ruled out, it had played a vital role in establishing the potential of repurposing activity during pandemic situation. An emergency use authorization (EUA) was granted to Remdesivir from the FDA on May 1, 2020, on the bases of its preliminary data.

With the advancement in technology and computational power, AI-facilitated drug repurposing with the help of powerful Machine Learning algorithms can prove beneficial in the COVID-19 scenario. Despite the availability of effectively affirmed repurposed drugs, there is still a need for finding more repurposed drugs (7).

Effectiveness of a Machine Learning algorithm is based on the availability of large amounts of data, which, fortunately is easily accessible because of the information discharged by the different health agencies and organizations on the open stages. Various types of ML models can be used for example supervised, unsupervised, and reinforcement models.

For drug repurposing, supervised models can be used to train classifiers based on the information available for similar conditions. Some of the examples are state-of-the-art machine learning approaches including deep neural networks (DNN), support vector machine (SVM), random forest (RF), gradient boosted machine with trees (GBM) and logistic regression with elastic net regularization (EN) to predict indications (8).

Few scientists have created a comprehensive biological knowledge graph relating genes, compounds, diseases, biological processes, side effects and symptoms termed Drug Repurposing Knowledge Graph (DRKG), based on a number of machine learning techniques (Ioannidis et al., 2020) (9)

Unsupervised learning models are also very common model used for drug repurposing activity.

I studied in an article a drug-centric unsupervised clustering approach for drug repositioning developed by integrating heterogeneous drug data profiles: drug-chemical, drug-protein and drug-side effect relationships and many other such relationships (10)

The methodology can be implemented in the deep learning algorithms for efficient data classification by developing graphs, based on the protein- ligand interaction.

Target identification and approval time for repurposed drugs are lesser with a 50-60% reduction in cost (11).

Healthcare industry is getting more and more competitive, albeit unpredictable, in the present-day scenario and there’s a need to be proactive, with faster, cheaper and more agile solutions. Although Artificial Intelligence usage has become very common in almost all the sectors, there’s a greater need to increase its usage in drug discovery and development, so that it can be used to overcome the limitations of discovery process.

I came across many interesting terminologies like deep learning, graphical neural networks, KNN algorithms (k-nearest neighbours’ algorithm), which were very intriguing. Their capabilities and scope in health care sector is astonishing

As a healthcare management student with a pharmacy background, I am well aware of the time constraints, limiting this research and development. Fortunately, I received an opportunity to explore this subject further, and to work on a project about the formulation of Machine Learning algorithms that aid in resolving various problems of the pharmaceutical R&D. Problems like prediction of the chemical properties of a molecule, building of an efficient classifier for screening of the drugs in repurposing activity, development of a powerful pharmacophore model etc. I chose the ten binding ligands for the most popular protein targets, 3Cl protease / m protease and papain like protease and started to work on the algorithm for a simple ligand beta mercaptoethanol. Similarly, more advanced and drug-able ligands can be used for making a Machine Learning models.

In my opinion, another interesting field is the pharmacophore model optimization that can yield amazingly accurate results as compared to manmade pharmacophore models. This is because the problem of human bias and errors can be decreased by a large margin.

Machine Learning has a tremendous scope in the field of pharmacy R&D and its benefits are still not fully enjoyed by the industry.   

REFERENCES

1. MARCH 27, 2020, H1N1 flu vs. COVID-19: Comparing pandemics and the response, by Blythe Bernhard
2. Applications of machine learning in drug discovery and development, Jessica Vamathevan,1,* Dominic Clark,1 Paul Czodrowski,2 Ian Dunham,3 Edgardo Ferran,1 George Lee,4 Bin Li,5 Anant Madabhushi,6,7 Parantu Shah,8 Michaela Spitzer,3 and Shanrong Zhao9
3. Building attention and edge message passing neural networks for bioactivity and physical–chemical property prediction, jan 2020, M. Withnall, E. Lindelöf, O. Engkvist & H. Chen
4. A self-attention based message passing neural network for predicting molecular lipophilicity and aqueous solubilityBowen Tang, Skyler T. Kramer, Meijuan Fang, Yingkun Qiu, Zhen Wu & Dong Xu, Journal of Cheminformatics volume 12, Article number: 15 (2020
5. Dawn Anderson et al., Digital R&D: Transforming the future of clinical development, Deloitte Insights, February 2018, accessed December 17, 2019.
6. Emergency Antiviral Drug Discovery During a Pandemic – a Case Study on the Application of Natural Compounds to Treat COVID-1,submitted on 15.05.2020, 03:44 and posted on 15.05.2020, 17:34 by Jianfeng Yu Shengxi Shao Bin Liu Zhihao Wang Yi-Zhou Jiang Yunqing Li Feng Chen Bing Liu
7. GNS H., Saraswathy G., Murahari M., Krishnamurthy M. An update on drug repurposing: re-written saga of the drug’s fate. Biomed Pharmacother. 2019;110(2):700–716. [PubMed] [Google Scholar]
8. A machine learning approach to drug repositioning based on drug expression profiles: Applications to schizophrenia and depression/anxiety disorders Kai Zhao1 and Hon-Cheong So*1,2 1 School of Biomedical Sciences, The Chinese University of Hong Kong, Shatin, Hong Kong 2 KIZ-CUHK Joint Laboratory of Bioresources and Molecular Research of Common Diseases, Kunming Zoology Institute of Zoology and The Chinese University of Hong Kong
9. Few-shot link prediction via graph neural networks for Covid-19 drug-repurposing Vassilis N. Ioannidis 1 Da Zheng 1 George Karypis 1
10. A two-tiered unsupervised clustering approach for drug repositioning through heterogeneous data integration, Pathima Nusrath Hameed, Karin Verspoor, Snezana Kusljic & Saman Halgamuge
11. Application of Artificial Intelligence in COVID-19 drug repurposing, Sweta Mohanty,a Md Harun AI Rashid,b Mayank Mridul,c Chandana Mohanty,a,∗ and Swati Swayamsiddha, Author information Article notes Copyright and License information Disclaimer.
12. Intelligent clinical trials Transforming through AI-enabled engagement; Karen, Taylor United Kingdom Francesca Properzi United Kingdom Maria Joao Cruz United Kingdom.
13. Stefan Harrer et al., Artificial Intelligence for Clinical Trial Design, ScienceDirect, August 2019, accessed December 18, 2019.
14. Giving Drugs a Second Chance: Overcoming Regulatory and Financial Hurdles in Repurposing Approved Drugs as Cancer Therapeutics J. Javier Hernandez,1,2,† Michael Pryszlak,1,3,† Lindsay Smith,1,3,† Connor Yanchus,1,2,† Naheed Kurji,4 Vijay M. Shahani,4 and Steven V. Molinski