We need a Big Data effort to find a COVID-19 cure, says pioneering geneticist

Famed French researcher Daniel Cohen, head of drug developer Pharnext, has a short list of 97 drugs that have been through some clinical testing that could play a role in a vaccine for COVID-19. To move things forward, the entire world needs to get together and share medical data on COVID-19 patients so the clues can be systematically studied.
Written by Tiernan Ray, Senior Contributing Writer

"I'm an optimist," says Daniel Cohen, a legendary French scientist, "If we can make this combined, coordinated effort, not just our laboratory, but all scientists working, worldwide, then, yes, I'm optimistic we can find a solution."

Cohen was talking about patients with COVID-19, the disease that results from the SARS-CoV2 virus, a disease that has engulfed the world's energy and attention. 

Cohen, 69, inaugurated the era of "Big Data" in the life sciences at the end of the 1980s, leading the effort then that produced the first "map" of the human genome.

As with the human genome project, where world leaders joined with scientists to generate funds for laboratories that shared information, Cohen thinks the world needs to share patient data from COVID-19 on a massive scale, including symptoms, medical history, and drugs the patients are on. 

"We must have a coordinated effort to mutualize COVID-19 medical records, including records of all the medications being given to patients," said Cohen, "to analyze, with artificial intelligence tools, the influence, positive or negative, of their conditions, and their drugs, on this disease."

Cohen spoke to ZDNet by phone on Sunday from Paris, where he is chief executive of development-stage drug maker Pharnext. The context was Cohen's announcement Monday that his Pharnext team has come up with 97 drugs that have been through various stages of clinical testing, that might play a role in an eventual vaccine for COVID-19. 

"Might" is the operative word. There is substantial complexity in the operation of the disease. It is widely known that primarily, but not exclusively, those dying are those who are older and who have "co-morbidities," pre-existing conditions that seem to become aggravated by the illness. The most prominent co-morbidities include hypertension, diabetes, cerebrovascular and cardiovascular diseases. 

There's a mystery as to why those conditions seem to make patients more susceptible. Is there an underlying "confounder," asks Cohen, a condition that produces both hypertension, say, and susceptibility to the virus. 

"Let's keep in mind that there are multiple potential other reasons for this comorbidity," says Cohen. "Organs from aged patients with diabetes or hypertension are often already damaged, therefore, COVID-19 might simply accelerate that degradation." 

PHARNEXT Janvier 2015

"I'm an optimist," says Pharnext chief executive Daniel Cohen. If scientists and hospitals around the world will pool data about patients with Covid-19, it will speed the way to use computers to refine a model of the malady and thereby infer which drugs may work in combination to stem the disease.

(Image: Arnaud Joron)

The notion that systems in the body express themselves in complex ways can be captured in a single term Cohen is fond of using, "pleiotropy," the notion that single gene, or any gene's protein, the "products" it makes, can have multiple, seemingly unrelated effects in the body.

Also: Graph theory suggests COVID-19 might be a 'small world' after all

For years, Cohen has been pursuing an approach that exploits pleiotropy, called "pleotherapy," which finds drugs to "re-purpose" that have shown efficacy in other diseases to fight new diseases for which they have not yet been approved. Often, the key is a combination of drugs. A Fortune Magazine article a year ago described his philosophy of drug development thusly: "My feeling is that with 50 drugs, we can treat everything."

Cohen's Pharnext used artificial intelligence, trained on large quantities of data, to arrive at a set of three drugs, baclofen, naltrexone, and sorbitol, that in combination with one another may be able to treat Charcot-Marie-Tooth disease, or "CMT," a rare neurological disorder for which no cure has been found, what's known as an orphaned disease.

That drug, called "PXT3003," is currently going through a second "Phase III" clinical trial to confirm the results from the first trial, said Cohen. The publication process for that is "extremely long but is going pretty well," said Cohen. A second drug is currently garnering "encouraging" results in Phase II trials for treatment of Alzheimer's.

The current work on COVID-19 has been posted on a pre-print server and bears the daunting title "Focusing on the Unfolded Protein Response and Autophagy Related Pathways to Reposition Common Approved Drugs against COVID-19." It is a fascinating piece of detective work. 

Also: MIT's deep learning found an antibiotic for a germ nothing else could kill

Cohen and his Pharnext colleagues started with a hypothesis. As always with pleiotropy, it's about looking more broadly at systems in the body, upstream from an illness, to find common mechanisms that might be functioning or misfunctioning. They focused on four "common pathways" in a disease that may be involved, meaning, chains of interactions of molecules. Viruses hijack the cellular signaling system of the body to replicate, so the heightened activity of the pathways can be a clue.

Cohen and the team went looking for proteins in COVID-19 that could be involved in accentuating those pathways. There isn't enough data gathered for SARS-CoV2, so they had to study the data from the original SARS-CoV virus, and from the MERS-CoV, the "Middle East Respiratory Syndrome" that broke out starting in 2012.

Searching the literature would have taken a dozen people multiple months to conduct, but was sped up by machine-learning software that let Cohen and a couple of colleagues complete it in two weeks, he said. 

With proteins and a hypothesis of the relevant pathways, they used a package from biotech company Qiagen that measures whether the pathways involving those viral proteins are "enriched" in the body. They did, indeed, find indications the four pathways they hypothesized are accentuated by the viral proteins.

With the pathway hunch confirmed, Cohen and his staff then scoured the literature again to see what drugs are involved in regulating those proteins. What they came up with is a shortlist of 97 drugs that not only work on the viral proteins but more broadly "modulate" those signaling pathways -- the multi-level effect one would expect to see with pleiotropy.

Those drugs span quite a large territory, including stent medication such as "rapamycin," and protease inhibitors for HIV treatment such as "lopinavir." Many of these, Cohen notes, are already involved in tests by drug makers for treatment of COVID-19, such as lopinavir, which was tested on patients in hospital in China. There is also the seemingly ubiquitous chloroquine, a remedy for malaria, which in its simple form, quinine, has been circulating as something of a folk remedy because it can be readily obtained by drinking tonic water.

The infatuation with quinine is no surprise to Cohen. If something familiar suddenly pops up as relevant, it's because pleiotropy implies complex ways that substances can act across the body and across maladies. Something in a malaria drug can touch something deeper that is common to both malaria and the virus.

As appealing as a gin and tonic would be in trying times, you shouldn't run out and buy these drugs for yourself and loved ones. The paper is not yet peer-reviewed, which means that its findings are not validated by other researchers. And Pharnext emphasizes that the list of drugs is not intended for self-medication. Any final drug that would come out of the list still involves a substantial amount of effort to produce something safe and effective for humans. The hope is that having a shortlist will speed up the process.

Cohen anticipates either partnering with a pharmaceutical company to develop those drugs, or Pharnext's developing them on its own, in collaboration with academic labs.

But first, the larger task, Cohen said, is the coordination of medical records that would allow the emerging evidence, currently scattered among different hospitals, to be studied systematically.

"This is called real-world evidence testing," Cohen told ZDNet. "It means that we have to adopt a strongly disciplined attitude, and establish homogeneous electronic records worldwide."

"This has to be started now, as soon as possible, with epidemiologists," he added. With Big Data, tools developed with machine learning forms of artificial intelligence can pull together all the complex ways in which pleiotropy makes one affliction turn into another.

It might be that the drugs patients are on for hypertension, say, have a role in suppressing the immune response, but it also might be that they can be marshaled in service of a stepped-up immune response if there is an underlying confounding condition that is common to both the comorbidity and to the virus. Moreover, the rule of pleotherapy, for Cohen, implies that any of these 97 drugs may function better as therapeutics if they are used in combination, not in isolation. Predicting those combined effects requires Big Data that can drive machine learning models that can peruse all possible combinations. 

Taking a step back, pleiotropy implies that COVID-19 and other coronavirus products might be telling a more complex and nuanced story about the world. The death rates, he notes, vary quite a bit from country to country, based on the data so far. Most patients are old and vulnerable, but there have also been pediatric cases. Looked at from one angle, the disease might simply be the tail end of a much longer chain of deleterious factors that afflict individuals broadly speaking. The disease, he reflected, "could be a consequence of aggravating factors linked to our modern society" in many forms, such as chemicals in the environment, stressors of various kinds. "We cannot exclude it," he said.

Embrace the complexity, is Cohen's prescription.

Editorial standards