This 2020 electron microscope image made available by the U.S. Centers for Disease Control and Prevention shows the spherical particles of the new coronavirus, colorized blue, from the first U.S. case of COVID-19. Antibody blood tests for the coronavirus could play a key role in deciding whether millions of Americans can safely return to work and school. But public health officials warn that the current “Wild West” of unregulated tests is creating confusion that could ultimately slow the path to recovery. (Hannah A. Bullock, Azaibi Tamin/CDC via AP)
Viral impression: A microscopic image from the first US case of Covid-19. The virions are coloured blue © AP

In late January, word disseminated through the medical world of an alarming new coronavirus disease that was spreading through China. In the UK, biotech company BenevolentAI turned its formidable artificial intelligence machine — set up to discover and develop new drugs — towards understanding the novel infection, then called 2019-nCoV.

The company used its “knowledge graph”, a large repository of medical information including connections extracted from scientific literature by machine learning, to look for existing medicines that could move quickly into clinical trials. AI enables it to solve pharmacological puzzles much faster than human experts, says Peter Richardson, BenevolentAI’s head of pharmacology.

“Rather than focusing solely on drugs that could affect the virus directly, we explored ways to inhibit the cellular processes that the virus uses to infect human cells,” says Joanna Shields, chief executive of BenevolentAI.

Within a week, a strong candidate had emerged: baricitinib, an oral drug marketed by Eli Lilly, the US pharmaceuticals group, to treat rheumatoid arthritis under the brand name Olumiant.

The analysis, published online in The Lancet medical journal on February 3 as the first cases of the new coronavirus were being reported in Europe, showed that baricitinib had several advantages as a treatment for Covid-19. Its anti-inflammatory properties could damp symptoms caused by a misdirected immune response in more advanced disease, while the drug also blocks viral replication by inhibiting an enzyme called AAK1 that the virus uses to enter human cells.

The AI exercise found further advantages of baricitinib over other drugs that target AAK1, says Richardson. Its side-effect profile is low at concentrations in blood plasma that are predicted to have an antiviral effect, and it appears to be compatible with direct-acting antivirals and HIV medications that are being tested as Covid-19 treatments. This opens up the possibility of combination therapies — where two or more drugs simultaneously target different aspects of Covid-19 pathology.

It was BenevolentAI that alerted Lilly to baricitinib anti-Covid potential. “Lilly responded with astonishing speed,” says Richardson. “They immediately went out and tested it in their labs. It did what we predicted.”

Lilly then scheduled a clinical trial of baricitinib in Covid-19 hospital patients in collaboration with the US National Institute for Allergy and Infectious Disease. The Indianapolis-based company says it plans to expand the baricitinib trial to hospitals in Europe and Asia, and expects results within two months.

Other AI-driven drug discovery and development companies are also directing resources to fight the pandemic. Exscientia, another UK biotech, is collaborating with Diamond Light Source, the country’s national synchrotron facility, and Calibr, the drug development division of Scripps Research in California, to develop compounds that could rapidly become viable treatments for Covid-19. Exscientia has access to Calibr’s 15,000 clinically ready molecules, which includes drugs already on the market and others that have passed clinical and animal safety studies.

Exscientia will first screen the complete molecule collection against key viral drug targets of Sars-Cov-2, the virus that causes Covid-19. Priority targets are enzymes that are vital for viral replication and also the virus’s spike surface protein, which interacts with the human cell receptor Ace2 to gain entry to human cells.

“The initial priority is to search for any existing drug that can be repurposed to protect the human population,” says Martin Redhead, head of quantitative pharmacology at Exscientia. “Then we can design superior molecules with our AI-design systems to work even more effectively against the virus.”

Protein puzzle: Diamond Light Source researcher Alice Douangamath aims X-rays at coronavirus proteins to see if they bind to various drugs
Protein puzzle: Diamond Light Source researcher Alice Douangamath aims X-rays at coronavirus proteins to see if they bind to various drugs

The role of Diamond — a facility in Oxfordshire that generates light beams 10bn times brighter than the sun — is to act as a molecular microscope, investigating how potential medicines interact with viral and human proteins and feeding the information back to Exscientia’s AI drug discovery algorithms. A third AI-powered UK biotech throwing resources into finding Covid-19 treatments is Healx, a Cambridge-based group looking for combination therapies.

Healx says that uncovering such treatments for Covid-19 requires detailed analysis of 8m possible pairs and 10.5bn triplets from the 4,000 approved drugs already on the market. Like Exscientia, the company’s AI platform uses a knowledge graph. Its version, called Healnet, integrates and analyses biomedical data from multiple sources to predict combinations most likely to succeed in the clinic.

“Our AI is able to combine two to three existing drugs to formulate the most effective treatment,” says David Brown, Healx chairman. “This approach ensures that any potential treatments we identify can be used by clinicians to help patients very quickly.” The company expects to have candidate combinations available in May for pre-clinical testing in collaboration with partners.

Beyond drug development, AI is being applied to analyse sounds and images of Covid-19. Researchers at Cambridge university have launched a mobile phone app that will collect data to develop machine learning algorithms for detecting whether someone is suffering from the disease — based on the sounds of their voice, breathing and coughing.

Because Covid-19 is a respiratory condition, it may affect these sounds in a specific way. The Cambridge team hope that a large, crowdsourced data set, collected using the app, can be used to develop machine learning algorithms for detecting the disease. The European Research Council is funding the project.

Cecilia Mascolo, the project leader in Cambridge’s computer science department, says that doctors have noticed the way patients with the virus catch their breath when speaking, as well as their breathing patterns and a dry cough.

“There are very few large data sets of respiratory sounds, so to make better algorithms that could be used for early detection, we need as many samples from as many participants as we can get. Even if we don’t get many positive cases of coronavirus, we could find links with other health conditions,” she says.

AI is becoming a useful tool for medical image analysis, and at least two separate UK groups are harnessing it to help diagnose Covid-19 symptoms in lung X-rays. Researchers at Birmingham City university have adapted a neural network called DeTraC (short for decompose, transfer and compose) to detect Covid-19.

Early experimental results showed that DeTraC could detect Covid-19 cases from an image data set collected from several hospitals around the world. It achieved 95 per cent accuracy in distinguishing Covid-19 X-rays from comparable images of other lung diseases.

Zegami, a data visualisation spinout from Oxford university, has developed a machine learning tool that aims to diagnose Covid-19 from lung X-rays. The company says it could not only differentiate Covid-19 from pneumonia caused by other pathogens, but could also predict likely outcomes based on the experience of previous patients.

To improve its AI tool, Zegami asked the UK’s National Health Service to provide it with data in the form of Covid-19 X-rays and details of patients’ treatment and outcome. “Covid-19 is a huge challenge and technology should play a key role in defeating it,” says Roger Noble, the company’s chief executive.

BenevolentAI’s Richardson agrees: “As human scientists working together we couldn’t possibly have identified baricitinib so quickly because there are so many competing ideas. We needed our AI ‘knowledge graph’ and the ability to query it to find new relationships and new ways to tackle disease.”

Copyright The Financial Times Limited 2024. All rights reserved.
Reuse this content (opens in new window) CommentsJump to comments section

Follow the topics in this article