Coronavirus: AI steps up in battle against Covid-19
Coronavirus: AI steps up in battle against COVID-19
It is as if the superhuman effort is needed to alleviate the global pandemic by killing so many.
Artificial intelligence may have been publicized, but when it comes to medicine, it already has a proven track record.
So can machine learning cope with this challenge of finding a cure for this terrible disease?
There is no shortage of companies trying to solve the dilemma.
Oxford-based Exscientia, the first to test an AI-discovered drug, is searching among the 15,000 drugs held by the Scripps research institute in California.
And Healx, a Cambridge company founded by Viagra co-inventor, dr. David Brown has revived his AI system developed to find drugs for rare diseases.
The system is divided into three parts which:
- trawling through all current literature related to the disease
- study the DNA and structure of the virus
- consider the suitability of various medications
The discovery of drugs has traditionally been slow.
“I have been doing this for 45 years and I have three drugs on the market,” said Dr. Brown to BBC News.
But AI is proving much faster.
“It took several weeks to gather all the data we need and in the last few days we have even gotten new information, so now we are at a critical mass,” said Dr. Brown.
“The algorithms worked for Easter and we will have the output for the three methods in the next seven days.”
Healx hopes to turn this information into a candidate drug list by May and is already in talks with labs to bring these predictions into clinical trials.
For those working in the field of AI drug discovery, there are two options when it comes to coronaviruses:
- find a completely new drug, but wait a couple of years for it to be approved as safe for use
- reuse existing drugs
But, said dr. Brown, it was extremely unlikely that a single drug could be the answer.
And for Healx, this means a detailed analysis of the eight million possible pairs and 10.5 billion combinations of three drugs derived from the 4,000 drugs approved on the market.
Professor Ara Darzi, director of the Institute of Global Health Innovation, at Imperial College, told BBC News: “Artificial intelligence remains one of our strongest pathways to achieving a perceptible solution, but there is a fundamental need large, clean, high-quality data sets.
“To date, much of this information has been silenced in individual companies such as large pharmaceutical companies or lost in intellectual property and old laboratory spaces within universities.
“Now more than ever, these diverse data sources on drug discovery need to be unified to allow AI researchers to apply their new machine learning techniques to generate new treatments for Covid-19 as soon as possible.”
In the United States, a partnership between the Barabasi Labs of Northeastern University, Harvard Medical School, the Stanford Network Science Institute and the biotechnological start-up Schipher Medicine is also looking for drugs that can be quickly re-proposed as Covid-19 treatments.
Normally, having them all work together would require “a year of paperwork,” said Schipher’s chief executive officer, Alif Saleh.
But a series of Zoom calls with a “group of people with unprecedented determination to do things, not to mention the time of their hands,” has speeded things up.
“The past three weeks would normally take a year and a half. Everyone has dropped everything,” he said.
Their research has already produced surprising results, including:
- the suggestion that the virus could invade brain tissue, which could explain why some people lose their sense of taste or smell)
- the prediction could also attack the reproductive system of men and women
Schipher Medicine combines AI with something called network medicine – a method that visualizes disease through complex interactions between molecular components.
“A phenotype of the disease is rarely due to the failure of a gene or protein on its own – nature is not that simple – but the result of a cascading effect in a network of interactions between different proteins,” Saleh said.
Using network medicine, AI and a fusion of the two led the consortium to identify 81 potential drugs that could help.
“Artificial intelligence can do a little better, not only by observing higher-order correlations, but also little independent information that traditional network medicine could lose,” said Prof. Albert-Laszlo Barabasi.
But AI alone wouldn’t work, they needed all three approaches.
“Different tools look at different perspectives but together they are very powerful,” he added.
Some AI companies are already claiming to have isolated drugs that could help.
BenevolentAI has identified Baricitinib, a drug already approved for the treatment of rheumatoid arthritis, as a potential treatment for preventing the virus that infects lung cells.
And it has now entered a controlled process with the American National Institute of Allergies and Infectious Diseases.
Meanwhile, scientists from South Korea and the United States who use in-depth learning to study the potential of commercially available antiviral drugs have suggested that atazanavir, used for the treatment of AIDS, could be a good candidate.
Other companies are using AI for other purposes, such as analyzing scans to ease the burden on radiologists and help predict which patients are most likely to need a ventilator.
The Chinese technology giant Alibaba, for example, has announced an algorithm capable of diagnosing cases in 20 seconds, with an accuracy of 96%.
But some experts warn that AI systems have likely been trained on advanced infection data, making them less effective in detecting the first signs of the virus.
The professor. Darzi had to make a global effort by policy makers to convince large pharmaceutical companies to join forces with small drug data stores, academics and research charities.
“Time has never been more important for drug discovery data to open its secrets for AI to assist in the battle against Covid-19,” he said.