. . World News – UK – Coronavirus sparked AI boom in healthcare. Was it worth it?

Welcome to WIRED UK. This website uses cookies to improve your experience and deliver personalized advertising. You can always unsubscribe or learn more by reading our cookie policy.

DeepMind’s scientific breakthrough is just the beginning: Thanks to the pandemic, AI is taking over health care at lightning speed

By
David Cox

In late January, scientists from DeepMind, Google’s London-based AI division, gathered to discuss whether they could do something about the brewery coronavirus pandemic. At the time, the spread of Covid-19 was still largely confined to the city of Wuhan, but as the number of cases continued to grow exponentially, machine learning experts from London to San Francisco were preparing to harness the power of AI to fight use the Sars-CoV-2 virus.

« Our first reaction was to see how we could help, » says Demis Hassabis, CEO and co-founder of DeepMind. “The focus was on our AlphaFold system, which we had shown that it can predict the 3D structure of proteins with unprecedented accuracy compared to other calculation methods. ”

In early March, DeepMind published predictions generated by AlphaFold for the structures of various SARS-CoV-2-associated proteins to expedite the process of understanding how the virus works. « Understanding the role of these proteins is critical to developing therapies for the disease, » says Hassabis. « The structure of one of these proteins has since been determined experimentally and is in good agreement with our predictions. It offers an insight into how tools like AlphaFold can better prepare us for a future pandemic. ”

On Monday, DeepMind announced that the same AI had made a giant leap forward. AlphaFold had solved a 50-year-old scientific puzzle, the “protein folding problem,” which involved determining the 3-D shape of a protein based on its amino acid sequence. This will pave the way for faster treatment and drug discovery development for nearly all diseases, including cancer, dementia and even infectious diseases like Covid-19.

By
Jack Needham

During the Covid-19 pandemic, predicting the protein structures of Sars-CoV-2 is just one way of using AI to cope with the growing emergency. In the past eight months, AI platforms have taken on an unprecedented role in healthcare as overwhelming patient stresses and staff shortages caused medical institutions to seek technology to fill the void.

From algorithms that try to optimize ventilator care by predicting which hospitals are most needed and when, to intelligent triage tools that collect information on symptoms, medical histories and recommendations from patients, that should be treated in the emergency room first. Machines have never had so much impact on patients’ lives.

« Covid-19 has been a tremendous accelerator for digital health as various technologies have been widely adopted that many believed would take a decade in a few months, » said Yonatan Amir, CEO of Israeli health technology company Diagnostics Robotics delivers its AI triage tool to healthcare facilities in the US, Israel and India and has recently signed extensive contracts with organizations such as the Mayo Clinic.

Within five weeks in March and April the Diagnostics Robotics 2 tools tested. 5 million patients. Overall, according to Amir, the demand for its technology in healthcare is 7. 5 times larger than before the pandemic. Such intelligent triage tools have made struggling hospitals both more effective and efficient. One example is Royal Bolton Hospital in the UK, which used a tool from Mumbai-based tech company Qure. ai to speed up the waiting time for the results of the chest x-rays.

« Clinicians receive reports immediately before a formal report is received from the radiologist, » says Shaista Meraj, consultant radiologist with the Bolton NHS Foundation Trust. “Doctors have 24/7 access to chest x-ray reports, which allows them to create quick management plans. ”

By
Gian Volpicelli

However, caution should still be exercised in introducing this technology. While AI systems can handle an intense workload and never overlook important details because they are distracted or tired, they can and still make mistakes in their predictions because of certain subtleties in interacting with a patient or in assessing their overall symptoms overlooked.

“When a doctor looks a patient in the eyes during a routine exam, he can tell whether the patient understands or not. AI doesn’t have these important nuances yet, ”says Amir. « One of the things we realized is that the accuracy of the AI ​​prediction is often related to how the patient question is designed. ”

Amir explains that the ability to rephrase a question in different ways so that the patient understands it is a skill that current AI systems do not have, but are likely to develop in the years to come. « We ran a test that found that most patients didn’t know how to answer the question when asked whether their headache came on suddenly or if it developed over time, » he says. « But when we asked the question in » Was it suddenly like a lightning strike on the head? « Changed, the answers changed completely. As the AI ​​learns more languages ​​and understands the subtleties of the nuances, it becomes smarter and more accurate with its triage. ”

The question of how much one can trust the predictions of an algorithm has also arisen in the area of ​​the conversion of existing drugs against Covid-19. Back in January, scientists at London-based tech startup Benevolent AI began using the company’s « knowledge graph » – a large database of medical information made up of compounds extracted from scientific literature through machine learning – to try to identify existing drugs trials that could be accelerated into clinical practice.

On 4. In February, the company published its analysis in the medical journal The Lancet. In particular, it was suggested that baricitinib, a small molecule approved for the treatment of rheumatoid arthritis, could be effective against Covid-19. Nine months later, the US Food and Drug Administration (FDA) granted emergency approval for baricitinib for the treatment of Covid-19 patients in hospital. A phase III study showed that the likelihood of a patient’s condition getting worse when using the drug was significantly lower.

By
Chris Baraniuk

Olly Oechsle, a senior software developer at Benevolent AI, says the rapid approval is « an important milestone, » adding that it « has been progressing at an unprecedented pace, moving from computer to bank to bedside in nine months « . .

Baricitinib may be an AI success story, but it’s also an outlier. Evelyne Bischof, a clinician and researcher at the University of Medicine and Health Sciences in Shanghai, says there are currently at least 81 studies using machine learning algorithms to recommend drugs that could be used for Covid-19, but no others have received clinical approval. « On the scientific side, AI has certainly been successful in developing and converting drugs for Covid-19, » she says.

“On the clinical side, we still see few concrete, applicable examples. « This is not necessarily AI’s fault. Disappointing results from antivirals like umifenovir and remdesevir, touted as potential treatments for Covid-19 at the start of the pandemic, have dampened doctors’ excitement for drug reuse somewhat. In addition, algorithms still require the tedious process of randomized control studies to validate their predictions. This means that it can take at least many months for a proposed drug to reach the wider patient population.

Scientists faced similar hurdles when trying to develop entirely new drugs against the viral Sars-CoV-2 proteins. While DeepMind’s AlphaFold platform correctly predicted the structures of a number of virus proteins, it is still under development. Hence, their predictions required months of laboratory testing before scientists believed they could actually act on the predictions.

At the same time, the scale of the pandemic has meant that other AI solutions have become mainstream simply out of necessity. Billions of doses of Covid-19 vaccines are slated to be administered worldwide in the next few years after advancing through clinical trials at breakneck speed. Medical regulators are concerned about being able to follow up on reports of safety concerns.

By
Sanjana Varghese

The UK Medicines and Health Products Regulatory Agency (MHRA) has predicted that for every 100 million doses injected, 50. 000 to 100. There may be 000 reports of suspicious side effects. New York-based AI company Genpact has been awarded a contract to develop a platform that uses natural language processing to extract and analyze adverse event reports submitted by doctors on the MHRA website.

AI, used on big data, could also play a role in identifying retrospective clues as to why the pandemic affected some people disproportionately than others, and in identifying patterns in data that would be beyond human capabilities. Life science company LabCorp, based in Burlington, USA, is using a similar technology to break down an anonymized patient record made up of thousands of handwritten notes from doctors on Covid-19 patients. The aim is to find trends that can provide answers to questions such as:. B.. why Covid-19 hits vulnerable socio-economic groups particularly hard.

If this proves successful in the coming months, reliance on such AI tools will increase and they will inevitably play an increasingly important role in medical care. And when the next pandemic hits, it could mean we can respond much faster and more effectively. « As transformative as a technology may be, this pandemic came a little too early for the current capabilities of AI, » says Hassabis. “If we were unlucky enough to find ourselves in this situation again, I firmly believe that AI could play a crucial role. ”

Demis Hassabis, CEO and co-founder of DeepMind, was a speaker at this year’s WIRED Live 2020. The conference on 24. November brought together disruptive minds from technology, design, art and politics to examine how innovation, technological progress and world events are changing the way we live.

🇨🇳 Much of the world is suffering from a second wave – but China is not. Find out how Covid-19

was destroyed here

📧 Nigerian Prince email fraud has a long legacy, including fueling the growth of a notorious police unit

🔊 Listen to the WIRED Podcast, The Week in Science, Technology, and Culture that airs every Friday

WIRED World in 2021 is an unmissable introduction to the ideas, trends, technologies, people and companies that will shape the coming year.

By
Maria Mellor

By
Will Bedingfield

By
Amy Hawkins

By
WIRED

DeepMind, Artificial Intelligence, Protein, Biology, Protein Folding