Original article can be found here (source): artificial intelligence
Drug discovery is a notoriously long, complex and expensive process requiring the concerted efforts of the world’s brightest minds. The complexity in understanding human physiology and molecular mechanisms is increasing with every new research paper published and for every new compound tested. As the world is facing a new challenge in trying to both adapt to and defend itself against the coronavirus, artificial intelligence is offering new hope that a cure might be developed faster than ever before.
In this article, we will present some of the technologies being developed and applied in today’s drug discovery process, working side-by-side with scientists tracking new findings, and assisting in the creation of new compounds and potential vaccines. In addition, we will examine how the industry is applying AI in the fight against the coronavirus.
AI Applications for Drug Discovery
Start-ups focusing on the use of artificial intelligence in drug development and clinical trials have seen significant investment in recent years, and vendors focusing specifically on drug design and discovery received the majority of the total $5.2B funding observed between 2012 and 2019
Information Engines are fundamental machines behind applications in both drug discovery and clinical trials, serving as the basic information aggregator and synthesizer layer, on which the other applications can draw their insights, conclusions and prescriptive functions. The information available to scientists today is increasing exponentially, so the purpose of information engines being developed today is to help scientists update and aggregate all this information and pull out the data most likely to be relevant for a specific study.
The types of information going into these engines vary broadly. An advanced information engine integrates information from multiple sources such as scientific research publications, medical records, doctors journals, biomedical information such as known drug targets, ligand information and disease-specific information, historical clinical trial data, patent information from molecules currently being investigated at global pharma companies, proprietary enterprise data from internal research studies at the individual pharma client, genomic sequencing data, radiology imaging data, cohort data and even other real-world evidence such as society and environmental data.
In a recent analyst insight, we discussed how these information engines are being applied in clinical trials to enhance success rates and reduce associated trial costs. When it comes to the upstream processes relating to drug discovery, their purpose is to synthesize and analyze these vast amounts of information to help the scientist understand disease mechanisms and select the most promising targets, drug candidates or biomarkers; or as we will see in the next section, to assist the drug design application in creating the molecular designs or optimize a compound with desired properties. Information is typically presented via a knowledge graph that visualizes the relationships between diseases, genes, drugs and other data points, which the researcher then uses for target identification, biomarker discovery or other research areas.
AI-based drug design applications are involved directly with the molecular structure of the drugs. They draw data and insights from information engines to help generate novel drug candidates, to validate or optimize drug candidates, or to repurpose existing drugs for new therapeutic areas.
For target identification, machine learning is used to predict potential disease targets, and an AI triage then typically orders targets based on chemical opportunity, safety and druggability and presents them ranked with most promising targets. This information is then fed into the drug design application which optimizes the compounds with desired properties before they are selected for synthesis. Experimental data from the selected compounds can then be fed back into the model to generate additional data for optimization.
For drug repurposing, existing drugs approved for specific therapeutic areas are compared against possible similar pathways and targets in alternative diseases, which creates an opportunity for additional revenue from already developed pharmaceuticals. It also gives potential relief for rare disease areas where developing a new compound wouldn’t be profitable. Additionally, keeping repurposing in mind during the development of a new drug as opposed to having a disease-specific mindset, may result in more profitable multi-purpose pharmaceuticals entering the market in the coming years.
AI in the Fight Against Coronavirus
Recent substantial investment in AI for drug development has meant the start-ups have had the manpower and resources to develop their technologies. Compared to AI in medical imaging the total investment has been more than four-fold, even though the number of funded start-ups is equivalent between the two industries. This makes the average deal size for AI in drug development 3.5 times bigger than in medical imaging. The funding has been spent on significantly expanding and building capacity, as the total number of employees across these AI start-ups is now close to 10,000 globally.
A strong focus for start-up vendors is to create tight partnerships with the pharma industry. For many still in the early product development stages, this gives them the ability to test and optimize their solutions and to create proof-of-concept as a basis for additional deals.
For the more established start-ups, partnerships with the pharmaceutical industry turn the initial investments into revenue in the form of subscription or consulting charges, and potential milestone payments for new drug candidates, preparing the company for further investments, IPO, acquisition or continued success as a separate company. Pharmaceutical companies with high numbers of publicly announced AI partnerships include AstraZeneca, GSK, Sanofi, Merck, Janssen, and Pfizer, but many more are actively pursuing such opportunities today.
Many AI start-ups are therefore in the phase where they have a solution ready and are either looking for further partnerships or would like to showcase their solution and capabilities. The COVID-19 pandemic has, therefore, come as an important test for many of these vendors, where they can demonstrate the value of their technologies and hopefully help the world get through this crisis faster.
Understanding the protein structures on the coronavirus capsule can form the basis of a drug or vaccine. Google Deepmind have been using their artificial intelligence engine to quickly predict the structure of six proteins linked to the coronavirus, and although they have not been experimentally verified, they may still contribute to the research ultimately leading to therapeutics.
Hong Kong-based Insilico Medicine took the next step in finding possible treatments, using their AI algorithms to design new molecules that could potentially limit the virus’s ability to replicate. Using existing data on the similar virus which caused the SARS outbreak in 2003, they published structures of six new molecules that could potentially treat COVID-19. Also, Germany-based Innoplexus has used its drug discovery information engine to design a novel molecule candidate with a high binding affinity to a target protein on the coronavirus while maintaining drug-likeness criteria such as bioavailability, absorption, toxicity, etc. Other AI players following similar strategies to identify new targets and molecules include Pepticom, Micar Innovation, Acellera, MAbSilico, InveniAI and Iktos, and further initiatives are announced daily.
It is important to remember that even if AI helps researchers identify targets and design new molecules faster, clinical testing and regulatory approval will still take about a year. So, while waiting for a vaccine or a new drug to be developed, other teams are looking at existing drugs on the market that could be repurposed to treat COVID-19. BenevolentAI used their machine learning-based information engine to search for already approved drugs that could block the infection process. After analyzing chemical properties, medical data and scientific literature they identified Baricitinib, typically used to treat moderate and severe rheumatoid arthritis, as a potential candidate to treat COVID-19. The theory is that the drug would prevent the virus from entering the cells by inhibiting endocytosis, and thereby in combination with antiviral drugs reduce viral infectivity and replication and prevent the inflammatory response which causes some of the COVID-19 symptoms.
But although a lot is happening in the industry right now and there are many suggestions as to what might work as a therapy for COVID-19, both from existing drugs already on the market and from new molecules being designed by the AI drug developers, the scientific and medical community, as well as regulators, will not neglect the scientific method. Suggestions and new ideas are essential for progress, but so is rigor in testing and validation of hypotheses. A systematic approach, fuelled by accelerated findings using AI and bright minds in collaboration, will lead to a better outcome.
About Dr. Ulrik Kristensen
Dr. Ulrik Kristensen is a Senior Market Analyst at Signify Research, an independent supplier of market intelligence and consultancy to the global healthcare technology industry. Ulrik is part of the Healthcare IT team and leads the research covering Drug Development, Oncology, and Genomics. Ulrik holds an MSc in Molecular Biology from Aarhus University and a Ph.D. from the University of Strasbourg.