AI Is Screening Thousands of Existing Drugs to Find Out What Works Against COVID-19

April 6, 2020 - 9 minutes read

As the novel coronavirus (COVID-19) continues to wreak havoc around the world, doctors and medical researchers are desperately looking for medications to combat the illness effectively. And artificial intelligence (AI) development may help point them in the right direction.

Deep neural networks could help healthcare providers identify the right antivirals to fight COVID-19. And they’re not just looking at experimental drugs for a solution — the algorithms behind these networks are also considering already-existing compounds.

The Frantic Search for a Solution to COVID-19

If you’ve been following the news about coronavirus, you’ve probably heard of chloroquine by now. Based on a naturally-occurring compound found in certain South African trees, this anti-malaria drug was created by German medical developers in the 1930s. Since then, it has been saving lives across the globe.

As a last resort, Chinese physicians tried using chloroquine (along with numerous other drugs) on patients with severe COVID-19 cases. Some of them recovered. Many of them didn’t. Basically, nobody is sure if it really helped. And we won’t be able to confirm so without clinical trials, which are ongoing.

Alongside chloroquine, several other existing drugs are being investigated for potential efficacy in combating the coronavirus. Currently, “there are no definitively effective drugs,” according to Dr. Li Haichao, a respiratory and critical care doctor at Peking University First Hospital. Dr. Haichao is also a member of China’s emergency medical rescue team which was sent to Wuhan.

While these drug candidates may range in application, they do share one common characteristic: None of them are new. None of them were developed specifically to treat coronavirus patients. But they each have traits that could potentially combat the illness.

Repurposing already available drugs is one of the quickest ways to treat an outbreak. Developing new drugs is not only daunting but time-consuming — the process can take a decade. Existing drugs that have already been approved by regulatory agencies can spring into action much faster and start saving lives.

Before AI, scientists would be left to make educated guesses as to what works in crises like this. But a recently released preprint research paper looks at how deep neural networks could help. Not only do their algorithms scan new compounds, but they also consider already-approved medications for effectiveness.

Drug Repurposing May Be Our Best Shot

The preprint is one effort of many that are using AI tools and machine learning applications for drug discovery. AI can help this process in numerous ways; identifying new targets, searching for novel molecules and finding compounds with potential to pass through clinical trials and make it to market name just a few avenues.

Most AI-fueled drug discovery endeavors concentrate on new compounds. But COVID-19’s rapid damage to global health and economies has urged researchers to consider drug repurposing as a promising option.

The concept of applying a drug for one illness to another sounds strange and nonsensical on the surface level. After all, if it took a decade to develop that drug to work against one sickness, why would it be effective for something else? Two words: Biological similarity.

COVID-19 may be new to humans, but it’s not exactly unique to evolution. Because COVID-19 is a type of coronavirus (and virus in general), we have some idea of how it infects cells and transmits based on our dealings with similar viruses like SARS and MERS. Going deeper, we can essentially match up how our bodies respond to this illness on a molecular level by comparing it to the precedent of these other viruses.

In more technical terms, if a drug has a similar effect on gene expression profiles between two different circumstances (two infections, in this case), then perhaps the drug can be applied to the new infection. At least, that’s the logic here. But from a practicing physician’s perspective, this logic may not carry as much merit.

Chloroquine may have exhibited antiviral properties on cells in an isolated lab experiment, but “no acute virus infection has been successfully treated by chloroquine in humans.” Its use on COVID-19 patients was really a desperate attempt to save them. In the past, it seemed to help against SARS. But this was never truly confirmed.

Besides this, familiarity can backfire. A drug that was approved for one reason may not be questioned in terms of safety when it’s applied for another reason. But that can be dangerous. For example, the difference between a therapeutic dose of chloroquine and a toxic, potentially life-threatening dose is extremely narrow.

AI’s Role in All of This

It’s worth noting that AI does have the capability to explore drug effects on the molecular or genetic level much better than a human doctor can. And if the puzzle pieces align, there may be promise there. If an HIV drug triggers the same gene expression changes in COVID-19 patients as it does with HIV, perhaps it could work.

In the preprint’s case, the researchers based their hypothesis on SARS, a virus that carries many similarities to COVID-19 — genetically speaking, they have an 86% similarity. So, in theory, a drug that successfully works against SARS could be promising for combating COVID-19.

This is where AI comes in. A gene called COPB2 was found to be essential in helping SARS proliferate in the human body. The research team examined the genetic profile of cells without this gene; they are at least partially SARS-resistant and could be COVID-19-resistant. The team then screened through various chemical libraries to identify compounds that would basically eliminate the COPB2 gene in cells.

The researchers’ neural network ended up with a list of both experimental and approved compounds that matched this profile. For example, one chemical on the list was indeed previously found to reduce SARS replication in cells.

Not a Panacea, but a Useful Tool in Our Fight Against COVID-19

If all of this has left you with more questions than answers, you’re not alone. We’re just beginning to understand COVID-19. That means there’s little data on it to train AI with. The research team used SARS as a proxy, which, when considering its similarity to COVID-19, is logical. But vital questions, like if COPB2 is actually necessary for COVID-19 to proliferate, still need to be answered.

Everyone is eager to find a solution to COVID-19. Who wouldn’t be? From our home in Los Angeles to Venice, Italy, this illness has taken away too much from us already.

But letting hope outweigh truth and data could be detrimental in our progress. We must remember that AI is a tool, not a panacea. And just because a drug has been approved for one disorder doesn’t mean it will be effective against another similar one. We can’t let scientific objectivity take a backseat — not if we want to solve this crisis properly.

AI can help us immensely. But it’s ultimately up to clinical trials to validate a drug’s effectiveness against COVID-19.

Tags: , , , , , , , , , , , , , , , , ,