Although many scientists and technology leaders including Bill Gates in his now infamous TED talk in 2015 warned of the likelihood of a highly infectious airborne virus potentially leaping species to infect humans on a mass scale, it seems that nearly every country in the world was caught off guard. As a result, a global lockdown of nearly 40% of the world's population became necessary whilst governments dusted off pandemic plans, established emergency supply chains for protective equipment and developed testing protocols in response to the COVID-19 global pandemic.
The sudden appearance and rapid global spread of the novel coronavirus illustrated the need to leverage emerging technologies to hunt down treatments to combat the virus until a vaccine could be developed, tested and available in the global supply chain. Following China’s publication on 12 January 2020 of the genetic sequence of COVID-19, the race was on to identify drugs that might have a prophylactic effect of reducing the severity of the virus or that could inhibit the virus and the body’s hyper-inflammatory response to it. Several antiviral drugs previously used to treat malaria and HIV were put forward by pharma companies, including an experimental drug developed for Ebola all of which started clinical trials.
In a global pandemic, speed is of the essence – and machine learning models excel in handling data in fast-changing circumstances. Computing power and algorithms can be harnessed to work as tireless and unbiased super-researchers, analysing chemical, biological and medical databases to identify potential drug leads far faster than humans. AI systems save time and the agnostic approach adopted by machine learning means such platforms can process all of the world’s available bio information to generate leads that may be overlooked by traditional research.
At BenevolentAI, a technology & drug discovery company based in London and Cambridge UK, we set up a specialist scientific team in late January and launched an investigation using our AI drug discovery platform to identify approved drugs which could potentially treat COVID-19. Using a vast, curated biomedical knowledge graph, we surfaced a number of potential drugs and through a triage process using deep learning models, BenevolentAI scientists identified baricitinib, an approved rheumatoid arthritis drug as a potential treatment. Although many scientists were already studying similar anti-inflammatory drugs, using the AI platform, we discovered that Baricitinib, an IL6 inhibitor known to reduce inflammation in the body, has anti-viral properties that could inhibit endocytosis, the process by which a virus infects cells as well as reducing the body’s extreme immune response referred to as a cytokine storm. Baricitinib is a small molecule, readily available in the supply chain, taken orally and renally cleared in twelve hours, meaning it can be rapidly put to work either alone or in combination with other therapies.
At the start of February BenevolentAI published research findings in The Lancet, which was followed by a second publication in Lancet Infectious Diseases and by March, small groups of Italian physicians battling the virus on the frontlines became aware of the Lancet publications and commenced treating COVID-19 patients with baricitinib on a compassionate basis. Those investigator-led trials yielded initial successful outcomes. Eli Lilly, which owns the drug baricitinib subsequently validated the hypothesis in vitro and on 10th April, announced it had entered into an agreement with the National Institute of Allergy and Infectious Diseases (NIAID) to test baricitinib in clinical trials and alongside Remdesvir as a treatment protocol.
Note: On April 29, 2020, the US Center for Disease Control reviewed early results of the remdesivir randomised clinical trial that showed the drug could improve time to recovery by approximately 30% and within days of releasing those findings, approved it for emergency use. Remdesivir is an injectable drug that must be administered by a medical professional in hospital and is not yet available in the global supply chain.
The incredible speed at which this hypothesis moved from computer to bench to bedside demonstrates the far-reaching potential of AI models and algorithms for identifying existing and new drugs for the treatment of disease. While the urgency of the coronavirus outbreak means it makes sense to hunt through already approved drugs that could be ready for large-scale trials within weeks, the greatest potential lies in uncovering brand-new treatments for complex diseases with poorly understood mechanisms of action that have defied conventional research efforts. Indeed, this is where BenevolentAI’s primary effort is focused; working on complex diseases that currently have no effective treatment such as Glioblastoma Multiforme, ALS, Crohn’s Disease and Ulcerative Colitis.
Tackling an infectious disease may have been a new avenue for this AI-enabled technology but it may not be the last since evidence suggests the current coronavirus outbreak could represent an increasingly frequent pattern of epidemics, fuelled by our hyper-connected modern world.
New tech, new challenges
Data is the lifeblood of AI-powered research. Machine learning models and algorithms can interrogate vast quantities of biomedical data but the quality of that data has a direct impact on the value of hypotheses generated, the molecules designed and the patients identified as potentially benefiting from treatment. COVID-19 is ultimately a novel problem, and therefore the difficulty has been in collecting and applying quality data in a rapid timeframe. Urgency can lead to data being collected haphazardly jeopardizing results and usefulness in decision making. In addition, the pandemic has made clear the importance of collaboration and data sharing between governments, academic researchers and businesses but this must be accomplished in a way that also protects patient anonymity.
Another challenge centres around the issue of trust. This lack of trust is due to the ‘black box’ problem, whereby machine learning technologies are rarely able to explain the patterns they find. For humans - who demand transparency, evidence and a clear understanding of rationale - this inability to explain decisions results in a lack of trust in AI, which delays adoption of this transformational technology. Scepticism also arises from the simple fact that because algorithms are created by humans they can occasionally be wrong. In both cases, building trust involves a greater symbiosis between human and machine in the form of sustained efforts to refine predictions and better train models so that scientists and researchers can better understand where predictions have come from.
The success of AI in drug discovery is therefore contingent upon access to large datasets and a collaborative mindset which in the highly competitive traditional biopharma culture is a challenge indeed. A 2018 survey by the Pistoia Alliance, 52% of respondents stated a lack of access to data as one of the main barriers to innovation. Fortunately, recent years have seen a growth in collaborations and the open availability of large numbers of datasets, reflecting a growing awareness of the power of AI and other advanced technologies to transform drug discovery.
These are challenging times but there are many reasons to be hopeful. COVID-19 has helped accelerate data-sharing agreements and encourage open publication of research results and provided a glimpse of the beginnings of a more open and adaptable R&D model that can accelerate the delivery of innovative and life-changing products to patients. Alongside dozens of other scientific organizations and businesses, BenevolentAI signed the Wellcome Trust Pledge, to ensure that this and other research findings relevant to the coronavirus outbreak are shared rapidly and openly. The UK government followed by launching a new alliance to sequence the genomes of SARS-CoV-2. Backed by a £20 million investment, the COVID-19 Genomics UK Consortium (COG-UK) is comprised of the NHS, Public Health Agencies, the Wellcome Sanger Institute and several academic institutions. And finally, there are, at last count, over 100 vaccine projects in development around the world facilitated by the better understanding of the genetic makeup of the virus and open tracking of its mutations to identify if different strains are emerging.
Priorities for a post-Covid-19 world: enhance and augment
AI should not be regarded as a silver bullet, be it in pandemic response, healthcare generally, or drug discovery more specifically. In these areas, AI alone is not enough, for once an AI platform or system has made a prediction or recommendation, it requires interpretation by human expertise and capabilities. The fundamental role of technology should be to augment scientists’ capabilities, be it in identifying novel targets, designing novel molecules, or repurposing existing drugs.
In the search for potential treatments for COVID-19, AI technology played a significant role in accelerating potential drug candidates, streamlining the triage process and enhancing the ability to query these results. However, it was experienced scientists who evaluated those recommendations and put forward the hypothesis.
In a connected world in which pathogens spread at unprecedented speed, advanced technologies like AI and machine learning can be weapons to fight back. But while AI is proving its worth in the battle against this devastating disease, ultimately it is the fusion of machine intelligence and human ingenuity that holds the key to unleashing the full potential of this new technology to combat disease.
