In September 2021 leading experts from the world of healthcare and beyond gathered (virtually) for WIRED Health:Tech, our annual conference that explores the technologies driving the future of patient care. Here are some of the remarkable stories shared on the day.
The creators of the groundbreaking mRNA Covid-19 vaccine were previously developing the technology for cancer patients. Now it’s going back to its roots.
In late 2020, Ozlem Tureci became famous as one of the co-founders of BioNTech, the pioneering biotechnology German startup which successfully produced a mRNA Covid-19 vaccine in less than a year after the start of the pandemic. It was, however, a breakthrough that almost didn’t happen. Pfizer, who eventually became BioNTech’s global partners on the development and production of the vaccine, initially rejected the German’s startup approach to work on an experimental mRNA vaccine against the coronavirus. “The first discussion was in January 2020, so at a time where no one really believed that this new virus would become a pandemic,” Tureci said. “And as soon as it became more visible that this would become a pandemic, our Pfizer colleagues joined us in our effort to generate a vaccine.”
Today, while BioNTech continues its effort to mass-vaccinate the world against Covid-19—they have recently signed a deal with South African biopharma Biovac to undertake the final part of the vaccine’s production—they are also exploring the application of its mRNA vaccine technology to other diseases. In July, they announced the development of a malaria vaccine, a disease which many other promising vaccine candidates have failed to treat. “mRNA has a couple of advantages,” Tureci said. “It's a very versatile technology and a technology by which one can engineer very fast individual vaccine candidates which feature different antigens and this is exactly what we want to leverage: assessing different immunogens of a malaria pathogen and test them with regard to the potency to induce immune responses.” Clinical testing will start by the end of 2022, in partnership with manufacturing sites in Rwanda and Senegal. “We see the huge potential of mRNA technologies to democratise access to health. The long-term goal is to build dedicated end-to-end productions for an enduring long-term manufacturing solutions in those regions,” Tureci says.
BioNTech’s research efforts are also focused on a different epidemic: cancer. The company is already in advanced stages of randomised clinical trials for experimental treatments for melanoma and colon cancer. “That challenge is much higher as compared to infectious disease vaccines,” Tureci says. “There are established tumours and the vaccine has the task to shrink them. In other words, the versions of our cancer vaccine are highly individualised vaccines. Here we address the fact that every cancer patient has a unique tumour. What we do is to identify the molecular fingerprint in the mutations of every cancer patient and engineer an on-demand cancer vaccine for this patient within a couple of weeks in order to offer a tailored treatment.”
Cancer vaccines, of course, have long been the focus of BioNTech’s research. Their cancer mRNA vaccine clinical trials started in 2012. According to Tureci, by January 2020 they had already treated several hundreds of cancer patients when their research efforts had to be diverted to the Covid-19 pandemic. For Tureci, who started her medical career as a cancer physician, the successful development of a cancer vaccine would represent the culmination of a decades-long journey which started out of desperation: “We learned every day that the armamentarium, which we have, in terms of standards of care, in particular for advanced cancers, is a limited one,” she said. “We saw at the same time the potential of the technology and the science to provide new ways of treating cancer patients. And we also saw that it took a very long time to bring those technologies to cancer patients. We understood that the only way was to take things into our own hands.”
A personalised-nutrition app that uses solid research from a number of studies, plus rafts of self-reported data from users, could now transform preventive medicine.
Tim Spector’s blood sugar control is bad. His blood fat control is also poor. This means that, after a meal, he’s prone to high spikes in his sugar and fat levels. This is bad news, as these responses trigger hunger and inflammation which, in the long term, can lead to weight gain and chronic disease. “The only thing that's holding me together is my microbiome, which is in the top 5 per cent of the population,” Spector, a professor of epidemiology at King’s College London, said. “Which is just as well because I've written books on it. “
This analysis was provided to Spector by his personalised nutrition app ZOE, which analyses people’s responses to food. The app uses data derived from a study of over 13000 twins, one of the largest studies of its kind. When conducting that study, one of the first ‘aha’ moments for Spector was how variable people’s responses are when eating identical foods “We saw an eight to ten- fold difference between people responding to eating a muffin, both in terms of their sugar, insulin, but also their fat responses,” Spector said. “The other ‘aha’ moment was that twins respond very differently to the same food.” In other words, genetics were not that important.
ZOE requires users to do initial home-tests for their microbiome, blood sugar and blood fat. The app then offers dietary recommendations, ranking the foods individuals should be prioritising and offering support from nutritionists. Every six months, users do a round of retesting to track progress.
“The outcomes we've had so far have been very exciting,” Spector said. “People who followed their personalised care plan after three months’ experience a nine-and-a-half pound [4.3kg] average weight loss. 83 per cent generally felt less hunger and 82 per cent had more energy, a finding we hadn't expected.”
The idea for ZOE came five years ago, when Spector was approached by two entrepreneurs, Jonathan Wolf and George Hadjigeorgiou, after a talk he gave about one of his books, “The Diet Myth”. Wolf and Hadjigeorgiou wanted to launch an personalised nutrition app that could harness technologies like machine learning, at-home blood testing and continuous glucose monitoring. “I told them, ‘I'm not interested unless you're really going to fund research. You've got to go and find a few million dollars to go and get this to do it properly’”, Spector recalled. “Amazingly, they did find the money.”
ZOE’s first study, PREDICT, one of the world’s largest nutrition studies, involved researchers from Stanford, Massachusetts General Hospital, Harvard, Tufts and King’s College London. “That study was just wrapping up when Covid hit,” Spector said. “We came up with the idea of converting the nutrition app, and the whole team behind it, to do something about Covid.”
After five days of development, ZOE released the Covid Symptom Tracker app on the first day of lockdown in the UK. Within 24 hours, it had reached one million downloads. “It crashed our server,” Spector said. By September 2021, the app had more than 4.5 million users. “This app exceeded all our expectations,” Spector said. “We were filling an unmet need. People weren't able to see their doctors, they weren't able to get to hospitals, they were told to just shut up and go away, no one was listening to them. We listened to them, we took all their symptoms.” The ZOE study was the first to report many of now known symptoms of coronavirus infections, such as loss of smell and taste, delirium in the elderly, skin rashes and others. “In the initial months, no one from the UK government contacted us other than indirectly trying to shut us down,” Spector says. “There was a lot of resistance from institutions to what we were doing, despite the enormous appeal to the public.” Eventually, the UK government’s Chief Scientific Officer, Patrick Vallance, who had supported the app from the outset, persuaded the Department of Health to include ZOE on their portfolio of tests. “Things did change, but they were pretty dodgy the first few months,” Spector says.
While the Covid study is still ongoing, the personalised nutrition app—which has only been available in the US so far—will be launched in the UK in 2022. Spector is also planning to expand its use to other large-scale studies in diseases like dementia and cancer. “It could really change the way we do preventive medicine and it could really reduce the need for GPs, like a triage service and for a minute fraction of the cost of traditional health care,” Spector said. “You can, by tracking your health on an app, give you an early warning device about what's going on, catching things much earlier than is currently the case. That has enormous potential.”
By combining algorithms with brain implants and data on the connections between motor skills and brain activity, it’s now possible to give back a lost voice.
Last year, Pancho, 35, got a second chance to talk again. Fifteen years ago, he was involved in a car accident which left him completely paralysed in the arms and legs, and also took away his ability to speak. All he could do to communicate was to move a stick, attached to his baseball cap, and type letters one by one, at a rate of about five per minute. All that changed, when, in 2019, Pancho joined an experimental clinical trial conducted by neurosurgeon Edward Chang at the University of California San Francisco
For the past decade, Chang had been studying the brain circuitry behind our ability to speak. “Our vocal tract consists of the larynx and the vocal apparatus, including the tongue, the teeth, the jaw, the lips, all of which move very precisely to create consonants, vowels and words,” Chang explained. “It's an extremely complex process. It's one of the most complex motor skills that we do, even though all of us do it. Without thinking too much about it, it requires co-ordination across 100 muscles. And it's very fast, it communicates 120 to 200 words per minute, for the average speaker. This is a really extraordinary ability.”
Using a technique called electrocorticography, which involves attaching electrodes on the brain to record brain activity, Chang has been studying the brain activity that occurs when we speak. Over the past years, he has studied closely the activity in the area of the brain linked to movements in the vocal tract, understanding how activity in each of these locations correlates to movements in the vocal tract. “We've been able to discover multiple aspects of how the things like the larynx are encoded, how the different parts of the lips and the jaw are coordinated to give movement,” Chang said. “It's a very complex interaction.”
Three years ago, Chang achieved a breakthrough, decoding the neural code for every consonant in English. Using that knowledge, Chang’s team set about to synthesise synthetic speech from brain activity of epilepsy patients, using a machine learning algorithm. “This was really the first time that we have been able to achieve intelligible speech from decoding directly from the brain activity,” Chang said. “It translates brain activity into simulated movements of the vocal tract. And from those movements, we generate synthesised audio speech.”
When Chang met Poncho, he wasn’t sure the trial would succeed. “After 15 years of not speaking, it was completely unclear whether his brain still had the substrate and ability and function to carry out speech,” Chang said. “But apparently it does. I'm not saying it's the same, but it certainly has the capacity.”
Chang’s team implanted a 120-channel electrode array inside the patient’s skill, covering the relevant part of the brain. During sessions, they connected a digital connector to the electrodes to translate the analogue brain activity into a digital signal that was read and translated into words by a machine learning algorithm. “An algorithm detects when he is starting to speak, then an algorithm classifies words according to rank probability, using a neural network decoding, and then applies a statistical model of the sequence of words, which works like an auto-correction,” Chang said. Initially, Pancho was trained on a vocabulary of 50 words and was able to conduct basic interactions, like responding to greetings or simple questions. The next step will be to make a new fully implantable prototype on his brain and to expand his vocabulary to about 1000 words. “Our main aim is really trying to restore social communication,” Chang said. “The ability to actually carry on a conversation with a loved one.”
Thanks to their mixtures of complex peptides found nowhere else in nature, sea-snail venom could be a source of new therapeutic drugs.
Mande Holford’s favourite animal is the venomous sea cone snail. These slow-moving creatures of the sea prey on small fish and marine worms. Typically, when an unsuspecting victim is nearby, the snail, hiding under the sand, will extend its tongue towards it, firing a harpoon filled with venom into its victim, paralysing it instantly. They then proceed to swallow the fish whole. “If this isn't innovation, then I don't know what is because there's no reason this slow-moving snail should be able to feed on the very fast fish.” Holford, a professor of Chemistry at City University of New York, says. “So when most people think of venom in venomous animals, they think of them as agents of fear. But when I think of them as agents of change and innovation.”
According to Holford, more 15 per cent of animals in nature are venomous—which amounts to more than 200,000 different species including jellyfish, spiders, scorpions and snakes. “Nature has repeated and used venom all throughout the tree of life.” Holford said. “Venom has been able to transform what would typically be a physical warfare into a biochemical warfare.” Holford compares them to chemical cluster bombs, complex chemical mixtures filled with, amongst other molecules, short chains of aminoacids called peptides. The venom of killer snails, for instance, includes anywhere between 50 to 250 unique peptides.
Holford studies how these peptides manipulate cellular physiology. “You've got this complex mixture of hundreds of things coming at you very fast and where do they go? They attack conserved physiological molecules in the body, blood, membranes, brain,” Holford said. “That means that they're attacking all of the strongholds that help to preserve and keep an organism vital. These are important systems that are being attacked by venom, which is why it can act so fast, and why it can be so potent.”
These properties—fast-acting, potency and high specificity—are also qualities researchers look for when developing new therapeutic drugs. “Our job as scientists is to figure out how to take those amazing compounds and turn them into forces of good,” Holford said. “It is a successful biochemical innovation that nature has done and that is very good at manipulating cellular communication. One thing that it can do is to shut off cellular signals, so when we try to take venom and turn it into a drug, we're trying to shut off signals that are malfunctioning.”
Holford’s lab is currently focusing on the search for venom peptides that can be manipulated to produce drugs that treat pain in cancer patients. “We try to identify new venom peptides and characterise how they function on the surface of cells,” Holford said. “These are molecular targets called iron channels and receptors. We then try to determine how manipulating the interactions at the surface can lead to manipulating malfunction signals in pain in cancer.”
If she’s successful in finding a venom-derived therapeutic for pain, however, she won’t be the first. “There is a snail drug on the market, unbeknownst to everybody. Most people aren't aware,” she said. The drug, called Prialt, can be used to treat chronic pain in patients with HIV and cancer. One of its advantages is that, unlike morphine, it’s not addictive. “Prialt does have one serious side effect in that it's not active in the peripheral system, so you can't use it in the bloodstream or pop it like a pill—it has to be delivered via spinal tap into the spinal cord where it's active,” Holford said. “So that limits its use. No one's signing up for spinal tap to manage pain.”
Holford’s already identified a series of venom compounds with the potential to treat pain and that can also be administered orally. During lockdown, however, Holford’s team couldn’t get out in the field to study snails in their habitat, so they came up with a creative solution. “We wanted to make mini venom glands and sort of stockpile them the way we did with beans and toilet paper during the shutdown,“ Holford said. “We have this sort of biobank of organoids, we would be able to manipulate them genetically, and understand more about how these genes work and how they function. We're trying to get at this question of how our genes are weaponized for both ecological and therapeutic advantage.”
The role of AI and supercomputing power will be an increasingly important factor in identifying new drugs and harnessing healthcare data across many specialties.
In July, US tech firm NVIDIA launched Cambridge-1, the UK’s fastest supercomputer, to help researchers and biomedical startups accelerate medical discovery. “Cambridge-1 is the perfect scientific instrument,” said Kimberly Powell, vice-president of healthcare at NVIDIA. “It's built using NVIDIA DGX A100 and has a full stack of AI and high-performance computing tools.” They partnered with institutions like AstraZeneca, GlaxoSmithKline, Guy’s and St Thomas' NHS Foundation Trust, King's College London and Oxford Nanopore Technologies.
At WIRED Health:Tech, Powell announced the expansion of Cambridge-1 partnership to Peptone, a UK startup working on a protein engineering operating system. Using machine learning techniques and computational physics, Peptone models unstructured proteins, which are molecules that don’t possess a stable shape, making the design of drugs that can react with them a challenging problem. “The recent breakthroughs resulting from AI and protein structure prediction and simulations are reshaping the way protein drugs can be designed,” Powell said. With Cambridge-1, Peptone will be able to “scale up their efforts in the use of generative models and molecular simulations to improve the design of antibodies that help treat inflammatory diseases like asthma and psoriasis.”
Powell also announced that UK startups could now apply for access to Cambridge-1 until the end of 2021, a programme she believes will “help speed and scale innovations to get to market faster and improve patient outcomes sooner.”
“Biology is being digitised and it's being digitised at every level,” she said. “Biology is an information science, where each level and our understanding gets richer from the levels above it and below it. But in these terms, they're too complex for humans. And we know that we need artificial intelligence to help us model, search and predict biological processes to really exponentially accelerate discovery.”
One of the processes being reshaped, according to Powell, is drug discovery, mostly thanks to new AI tools. “The latest breakthroughs are in a class called self-supervised learning. It's using these new variants of network architectures called auto encoders, transformers and graph neural networks,” she said. “These new AI tools are really transcending the critical limitations of supervised learning, like convolutional neural networks that require hard to come by, especially in healthcare large label data sets.”
Extracting knowledge from healthcare data—from images to health records—will be vital to the process of drug discovery. According to Powell, about 80 per cent of health information is contained in doctor’s notes. For instance, using a technique called relationship extraction, these algorithms can read volumes of doctor’s notes, identifying links between diagnoses and prescribed drugs. The University of Florida is using NVIDIA’s nature language processing algorithms to do just that. Using a training dataset that includes more than 10 years of anonymised data from two million patients from specialities like oncology, internal medicine and clinical care, researchers at NVIDIA have developed what they describe as the “largest clinical language model in the world.”
“It truly is elucidating doctor's notes,” Powell says. “And it can be used for tasks to find patients for clinical trials, predict life threatening diseases, and give clinical decisions important support for doctors.”
From combating Covid misinformation by elevating trusted sources, to harnessing the power of AI in diagnoses, Google’s data-led approach to healthcare is yielding results.
On 29 March, during a government Covid press conference, one of the UK government’s scientific advisers displayed a graph showing how the public was responding to the lockdown. It showed that people’s travel to locations like work, supermarkets, parks and other recreational spaces had reduced drastically in response to the introduction of new social distancing measures. “That was very useful for understanding the sort of social and economic consequences of different kinds of policy,” said Alan Karthikesalingam, a research lead at Google Health and part of the team behind the data analysis.
“One of the really interesting aspects about the pandemic has been the role of large technology companies in the midst of all of this,” he said.“ The first thing was making a responsible effort to partner with trusted health care agencies around the world sources of authoritative and trustworthy information.”
Combating Covid-19 misinformation was one of Google’s top priorities, partnering with government health care agencies around the world to elevate content from trusted institutions like the WHO on YouTube, as well as providing information on local Covid testing centres in hundreds of languages around the world. “I think it shows the power of these kinds of everyday technologies, for empowering all of us with trusted information so that we can make better decisions about our health,” Karthikesalingam said.
During the pandemic, researchers at Google Health also found how Search trends data could be helpful in predicting the spread of the disease. “For example, searches relating to anosmia, the loss of sense of smell, spiked initially after it was reported that it appeared to be a symptom associated with Covid-19, but then also rose in regions just before increasing infection rates were then seen in those places.”
In another project, the Google Cloud team partnered with Harvard's Global Health Institute to build a Covid-19 public forecasting model, which can predict cases, hospitalisations, and deaths 14 days out, using public data. “The ability to extract this kind of information, make it anonymous and privacy preserving, and then make it available to public health agencies around the world who are trying to model the spread of coronavirus has been really important,” Karthikesalingam said.
The pandemic has given new impetus to find new ways to deliver healthcare to patients. “As a practising physician, it's been very obvious how, in the delivery of healthcare, we've had to find new ways of working,” Karthikesalingam said. “Even before the pandemic, there were already real challenges in making expert diagnostic tests with the best possible levels of accuracy and availability around the world.”
In India, for instance, there’s a shortage of 127,000 eye specialists. As a result, about 45 per cent of patients suffer vision loss before even receiving a diagnosis. “There you can see how these profound shortages of ophthalmologists and specialists are associated with very real health challenges, including preventable loss of vision, which has all sorts of knock-on consequences for quality of life,” he said. Using machine learning, Google has developed new screening tools for preventable blindness diseases like diabetic retinopathy and glaucoma. Currently these systems are as accurate as generalist in diagnosing disease. “The interesting thing about these machine learning models is really their ability to learn from example and learn from data and to improve over time, he said.
In London, Google is developing similar AI tools for mammography and breast cancer screening. “We were able to train these kinds of systems, not only to identify referable retinopathy but to also identify breast cancer in screening mammograms,” he said. The system that we developed appears to be very accurate with approximating the level of specialist screening services.”
- Why bosses are inflexible about flexible working
- Humans have broken a fundamental law of the ocean
- The world needs to crack battery recycling, fast
- Body language pseudoscience is flourishing on YouTube
- TikTok’s next big move? To become Facebook
- Norway is running out of gas-guzzling cars to tax
- Locked out of ‘God Mode,’ runners are hacking their treadmills
- Subscribe to the WIRED Podcast. New episodes every Friday
This article was originally published by WIRED UK
