Category: Technology

Machine learning came along at just the right time. The world is now awash in more data than ever before, and computer algorithms that can learn and improve as they perform data analysis promise to help scientists handle that information overload.

Yet researchers who think that machine learning by itself can help solve complex problems in science, engineering, and medicine, should strive for a more balanced approach, says Roman Grigoriev, part of a School of Physics team with new research suggesting a hybrid approach for conducting science that blends new era technologies, old school experimentation, and theoretical analysis. The research suggests faster solutions to complex, data-intensive riddles involving such issues as cancer, earthquakes, weather forecasts, and climate change.

“It’s a combination of existing theoretical understanding — as well as experimental data with machine learning,” says Grigoriev, Physics professor and lead investigator of the Dynamics and Control Group. “Oftentimes people who do machine learning kind of forget about theoretical understanding and almost rely totally on data. It’s relatively simple, but when there’s a lot of data and not enough structure in that data, that approach is bound to fail.” Grigoriev explains that there’s often just too much data to meaningfully analyze, at which point “the problem becomes intractable. Essentially, harnessing appropriate domain knowledge is critical for finding structure in the data.”

“Robust learning from noisy, incomplete, high-dimensional experimental data via physically constrained symbolic regression,” was in Nature Communications. Fellow School of Physics researchers involved in the study are Michael Schatz, professor and the School’s interim chair; graduate research assistant Logan Kageorge; and former graduate research assistant Patrick A.K. Reinbold.

The problem with high-dimensional data

Machine learning uses computer algorithms to find patterns in data, but “most popular machine learning approaches present results in a form that is hard to interpret and explain,” Grigoriev says. “Unless you understand the how and the why you can’t really say you understand a problem.”

Understanding and predicting complicated behaviours — by crunching a lot of dense, rich data — can help with fundamental and practical problems in science areas like weather forecasting and characterizing cardiac arrhythmias. The problem is that most of those arenas involve “high-dimensional” data, which means exactly what it sounds like: data with a lot of dimensions or variables, sometimes millions of them.

The dimensionality of the data is so large that “you get lost and it’s hard to see any trends,” Grigoriev says.

His team has come up with a hybrid approach that blends machine learning with elements of the traditional process of scientific discovery. That means a theoretical description, observations, designing experiments to test the description, and “then going back and forth between improving the theories, and designing new experiments. That’s been the traditional approach for hundreds of years.”

The foundation of Science’s understanding and progress relies on that scientific method — the combination of theory and experimentation. “They’re not developed just based on the data. They are developed using both existing knowledges as well as some general fundamental laws.”

An approach that spotlights the beauty of equations

Constraining the data to include just those variables that pertain directly to the experiment in question is vital in working with high-dimensional data, Grigoriev says.

“What this approach allows you to do is identify a simpler model that uses the variables you need. It’s a simplified description that applies to a particular situation but obtained using data that’s computational or experimental. It can do both.”

The result is represented in a mathematical model, Grigoriev says, and “once you see those equations, you understand what the variables are. The equations certainly help explain the essence of a physical problem.” His team’s approach was validated in the research with a fluid dynamics experiment. A thin layer of liquid was suspended in a rectangular tank, with magnetic and electrical fields shot through it to create what physicists call a turbulent flow — irregular shifts happening within the fluid layer that can rapidly change direction and magnitude.

Grigoriev and his team used their hybrid approach to analyze the accessible data, in this case the velocity of the water. Subsequently, they were able to reconstruct variables that couldn’t be measured directly, like water pressure and force.

This is the beauty of the equations — how much they allow you to do, Grigoriev says.

“What we do get is an equation, or set of equations, which are in a familiar form. We know how to explain, how to solve the problem using these equations. This is the nice thing about this approach. We’re working with variables whose meaning we understand; we know how to interpret them.”

The team believes the study’s results will lead to advances like faster, more accurate ways to make predictions of complicated behaviour in those large, real-world problems in science, engineering, and medicine. For example, as Grigoriev’s team’s research states, “the ability to identify and quantify important patterns and sequences in atmospheric turbulence should enable weather forecasts that are better and more rapid than those currently possible today.”

Source: Georgia Tech

The SARS-CoV-2 virus contains a gene that codes for a strange protein that could be a good target for drugs to fight COVID-19 and possibly other coronavirus infections, according to a new study from the University of California, Berkeley.

When the virus injects its genome into a cell, the so-called ORF3a gene is expressed to make a protein that moves to the surface of the cell and looks like an ion channel — a passage all cells have for ions like calcium to move in and out as they communicate with other cells.

But UC Berkeley researchers, who began investigating the protein’s structure with cryogenic electron microscopy (cryoEM) in January of last year, immediately after the pandemic began, discovered that the protein is nothing like other ion channels known to science. For one thing, it’s half a channel — it only pierces the cell membrane halfway. It also has an unusual fold not seen in other ion channels.

While no one knows why the virus carries a gene to make an ion channel — and a strange one, at that — researchers have found that by knocking out the ORF3a gene in SARS-CoV-2 and a related ORF3a gene in the original SARS virus, SARS-CoV-1, the severity of the disease is reduced, at least in animal models. This, the researchers say, makes it a good target for drugs to reduce the severity of human coronavirus infections. A high-resolution cryoEM structure published by the UC Berkeley team provides key information needed to find such drugs.

“It wasn’t clear how conserved this protein was among coronaviruses, but once we solved the structure, we could look to see if other genes in other coronaviruses are likely to adopt the same fold,” said Stephen Brohawn, one of the senior authors of the study and a UC Berkeley assistant professor of molecular and cell biology. “And it turns out that in all of the coronaviruses that circulate among bats and can infect humans, these 3a proteins exist. So, it could be an even broader possible target than we had thought at the beginning of the project.”

Brohawn and his UC Berkeley colleagues have already identified a drug that blocks the ion channel — though it also blocks other ion channels — and have identified mutations in the ORF3a gene that alter channel function.

“This shows, in principle, that one can find molecules that block this activity,” Brohawn said. “Now one of the goals that we have is to screen for small molecules that block the channel and that are specific for blocking 3a compared to other ion channels.”

High-resolution cryoEM structures

The team’s preliminary results were posted on the open-access preprint server bioRxiv in June of 2020, and the final results — with a much higher resolution cryoEM structure for the 3a protein — were published in the journal Nature Structural & Molecular Biology. The original online paper drew attention from those using computational approaches to find drugs to fight COVID-19, but it also helped improve a computer algorithm — Google DeepMind’s AlphaFold — that predicts a protein’s 3D structure from its amino acid sequence.

Brohawn, who is a member of UC Berkeley’s Helen Wills Neuroscience Institute, uses cryoEM to study the structures of ion channels in neurons. For this project, he teamed up with the lab of Diana Bautista, UC Berkeley professor of molecular and cell biology, who uses other methods to study ion channels, specifically those involved in chronic itch, pain and inflammatory disease. Together, the two labs provided three different pieces of evidence that the 3a protein does act as an ion channel. Though they demonstrated this in artificially-made lipid membranes called liposomes, these membranes are chemically similar to the membranes of cells, so presumably, the proteins also work as ion channels in coronavirus-infected cells.

The odd structure of the 3a protein — specifically, its blind pore, which is totally unlike typical ion channels that contain a tunnel that goes completely through the membrane to allow ions easy passage in and out — suggests that ions may instead diffuse down grooves along the outside of the protein.

According to Brohawn, the 3a protein is the smallest membrane protein channel that has been imaged by cryoEM, which has rapidly become the best way to determine the 3D structure of molecules, down to the level of individual atoms and the water molecules that surround them. Brohawn continues to research the 3a protein and two other potential ion channels — called E and 8a —in the SARS-CoV-2 genome, both to understand how they work and to find potential drugs to inactivate them.

Bautista continues investigating how the virus uses these ion channels to take over human and animal epithelial cells and neurons. Many viruses interfere with calcium signalling in cells, which may help them take over the cells’ molecular machinery and force them to make millions of copies of the virus instead of daily cellular housekeeping.

“We thought 3a was the most understudied and potentially the most interesting of these proteins and something we could probably make a dent in, if we jumped on it right away,” Brohawn said. “We were in a good position to move quickly on this, and it was really exciting to see it come together that fast.”

Source: UC Berkeley

When the H1N1 swine flu outbreak hit in 2009 it caused typical flu symptoms, but severe cases led to pneumonia and lung failure.

Doctors were unsure which treatments would work, but steroids, which dampen down inflammation, seemed like a good bet. Clinicians in intensive care units decided to set up a trial to see if steroids helped patients severely ill with swine flu.

But this proved impossible. ‘The pandemic came and went and we missed an opportunity to test whether even a simple intervention like steroids worked or not,’ said Professor Alistair Nichol, ICU doctor in St Vincent’s University Hospital in Dublin, Ireland, who worked in an Australian hospital during the 2009 pandemic.

‘There was a peak in intensive care unit (ICU) transmissions and some clinicians got frustrated in trying to set up a drug trial (that never happened),’ recalled Dr Lennie Derde, an intensive care doctor at UMC Utrecht in the Netherlands.

That pandemic killed tens of thousands of people, but ICU clinicians expected worse in the future. ‘This one was moderately bad,’ recalled ICU physician Professor Steve Webb at the Royal Perth Hospital in Australia, ‘but if we had a really bad one, we were just woefully unfit to be able to do drug trials.’ So ICU clinicians came together to be better prepared. ‘Nobody doubted there would be a next time,’ said Prof. Nichol.

Community-acquired pneumonia

Typically, a viral pandemic begins with a spike in unusual pneumonia cases in ICUs.

In March 2020, patients flooded ICUs with SARS-CoV-2 infections. This time, some ICU clinicians were ready from the get-go. A number of coronavirus patients were quickly recruited into an existing clinical trial calledREMAP-CAP, set up in the wake of the 2009 swine flu outbreak, to test which drugs worked on their pneumonia.

It was ‘set up in peacetime,’ explained Prof. Nichol, ‘so as to be ready for wartime if a pandemic was to arrive.’

‘We were able to flick the switch to include pandemic patients,’ said Dr Derde. ‘That is why we were able to include our first patient on the 9th of March.’

The trial tests various drugs – rather than the usual one or two – for community-acquired pneumonia, with all the regulatory and ethical approvals in place. It was set up by a project called PREPARE for patients suffering from severe community-acquired pneumonia caused by bacteria or viruses, which kills about one in five people with it in ICU, although the figures can be much higher.

Results on steroids from the trial contributed to the World Health Organisation recommending them for COVID-19 patients in September 2020. By then, REMAP-CAP, which now involves about 300 hospitals, was testing lots of interventions (it currently has 31).

Doing this sets it apart from a traditional randomised controlled trial, which is the gold standard to prove whether a drug is effective or not. Patients are usually given either one or two drugs or a placebo, and then outcomes are compared. REMAP-CAP is different – and more flexible – in that it tests many types of treatments at the same time. One patient can receive multiple interventions, which is not unusual for someone severely ill. A regimen of treatments is ‘much more in line with your typical clinical treatment,’ said Dr Derde, who is the European coordinator of the trial.

The clinical trial design means that ‘we can analyse the interactions between drugs, which is a huge advantage in a pandemic,’ she said.

‘A single Covid patient can be randomised for up to eight separate aspects of their treatment,’ said Prof. Webb. ‘We’re obviously learning much more quickly, because we’re testing so many different things simultaneously.’ A patient is enrolled in one treatment ‘domain’, and then is randomly assigned to one of a handful of interventions in that domain.

The 12 different ‘domains’ of treatment include anti-coagulation treatments, anti-inflammatory drugs, and immune modulating drugs. ‘One patient might be involved in six or seven different domains of the trial,’ said Prof. Nichol. In fact, this better reflects how a patient with COVID-19 is treated in hospital, according to the doctors.

Monoclonal

A big result – presented in a pre-print study, so not yet peer reviewed – is that two monoclonal antibodies (tocilizumab and sarilumab) reduced death from COVID-19 in severely ill patients, and time spent in ICU. Those benefits seem to be on top of those from steroids, says Prof. Webb.

Tocilizumab and sarilumab block a chemical signal called interleukin-6 (IL-6), which stokes inflammation. In severe COVID-19, the inflammatory response of a patient’s immune system begins to damage body tissue, in friendly fire, while attacking the virus. There was therefore good reason to think the drugs might be effective.

Indeed, once a new disease turns up, and its mechanisms are reported, an array of medications will be considered by physicians. But they do not know which ones work. The results for COVID-19, explains Prof. Webb, ‘was quite a scattergun approach of clinicians using repurposed medicines.’

Sometimes, as in the case of the malaria drug hydroxychloroquine, a small study and lots of hype persuades some that an old drug is worth a try. But it requires a large clinical trial, such as REMAP-CAP, which involves 6,000 patients, to show whether that drug is safe and effective. Physicians can now prescribe the IL-6 blockers and steroids with greater confidence, while hydroxychloroquine is left on the shelf.

Indeed, finding that a drug does not work, or causes a negative effect, can be equally beneficial. ‘The anti-virals and the anti-coagulants have not just been ineffective, but may result in worse outcomes for patients,’ said Prof. Webb, about recent preliminary findings. ‘It is just as important to identify treatments that are harmful.’

It is difficult for a physician to know whether a drug is helping or hindering their patients. ‘For an individual doctor at a patient’s bedside, it can be hard to tease out if giving someone something helped or if they would have gotten better anyway,’ explained Prof. Nichol. Severely ill patients usually receive multiple interventions, which makes knowing which treatment was effective tougher.

Combinations

The REMAP-CAP trial was created to be multifactorial, meaning that the effects of many different treatment combinations on patients can show up in the data. Bayesian statistics, which applies probabilities to statistical problems, makes sense of all the data to answer crucial questions, such as when a treatment reaches a positive result, or if it is shown not to work, or if there are positive interactions between drugs. Such a strategy with multiple treatments had been used for cancer, but never a global ICU study.

At the heart of these trials is a simple fact: it is impossible to guess which treatments will work. Small studies can be biased and inconclusive, which is why large randomised trials are needed.

‘Years ago, I used to think that there were things that definitely would work, and then they didn’t,’ said Prof. Webb. ‘I’ve been doing clinical trials for so long now that the only thing I am confident of is that, when you do trials, you will get surprises.’

Dr Derde praises the European Union’s actions in 2014 ‘because they funded the European part of what is now REMAP-CAP, with the vision that they needed to set up an infrastructure for pandemic research.’ Other funders around the world, from Australia to New Zealand to Canada and the US, subsequently joined the effort.

REMAP-CAP is an international effort now, with 300 hospitals taking part in about 20 countries, testing out 31 different interventions. The trial will continue to prove which treatments work in severely ill COVID-19 patients. As the virus surges in many countries and until the vaccine rollout can control the pandemic, proven drugs are needed to save lives.