Affordable, Fast, and Personalized: The Future of AI-Fueled Drug Discovery

August 20, 2020 - 8 minutes read

Drug discovery is a ruthless field in which most drugs take almost a decade to develop and, even then, they can reach the last stage of experimentation only to falter. About 90% of all drug possibilities fail. Drug research can cost anywhere from $2.5 billion to $12 billion per drug. With this low success ratio and high cost, it’s worth letting a computer step in and fail fast instead of funding a team of human researchers to fail slowly.

With artificial intelligence (AI) applications, drug discovery can advance faster, allowing researchers to experiment with only the most promising drug possibilities. With machine learning, 50 lab scientists can accomplish what it would take 5,000 researchers to do in the same amount of time. AI enables drug research and development to become faster, lower-cost, and more personalized.

AI’s Deep Dive Into Drug Discovery

No, we’re not talking about a Breaking Bad-esque AI robot literally making drugs for humans to test. We’re referring to an AI technique called generative adversarial networks (GANs) that utilize two neural networks to “fight” against each other in generating new outcomes. GANs were initially used for tasks like designing new objects or developing fake (but realistic) human faces.

But in 2012, Alex Zhavoronkov, a former computer scientist interested in biophysics, noticed that AI was getting really good at text, image, and voice recognition. Given enough data, which was not a problem at that time, it was easy to train an AI application. He saw that pharmacology had a breadth of large datasets. Zhavoronkov came up with the idea that he could probably use AI to speed up drug discovery. Using GANs, he wanted to let researchers describe drug features, like “The new drug should use concentration X to inhibit protein Y with minimal side effects for humans.” Then, the AI would run the possible drug compounds against these criteria.

Eventually, Zhavoronkov founded Insilico Medicine at Johns Hopkins University and got to work on implementing his theoretical AI-enabled medical application. The drug discovery engine at Insilico first reads through millions of data samples to find a signature pattern in a specific disease’s biological characteristics. Then the engine flags the most promising treatments. Using GANs, the engine creates molecules that fit the disease’s criteria for a successful drug. The result, Zhavoronkov says, “is an explosion in potential drug targets and a much more efficient testing process.”

In 2018, Insilico took 46 days to generate new molecules, and this included the discovery, drug synthesis, and computer simulations of experiments. Now, it takes a month or two to identify a promising drug, rather than a decade. The company is focused on several major diseases to fight: fibrosis, Alzheimer’s, Parkinson’s, cancer, aging, diabetes, ALS, and more.

By the end of 2020, the first drug will reach Phase I trials, which aims to narrow down the drug dose to an amount with the fewest side effects. The drug in question would be a treatment for hair loss. The company is also starting to use AI to predict clinical trial outcomes before the trials occur. If this part of the engine is successful, it’ll cut down even more time in the drug discovery and experimentation process.

Protein Folding Permutations

AI is also being applied by other researchers to find new drug targets. Drugs need to bind to proteins and molecules in the body to start working. This part of the drug discovery process is incredibly important; if a drug cannot bind, it is completely useless to the human body.

Because of the complexity and possibilities of a 3D protein’s folding structure, researchers have only managed to discover five new drugs per year between 1980 and 2006. Even with the annual investment of $30 billion, research output hasn’t improved in quantity or quality. A protein with only a hundred amino acids (this is considered pretty small) can produce more than 10100 potential folding shapes.

Even the world’s best supercomputers have trouble with generating all of the possible protein foldings. In 1994, a competition was created to monitor the progress of the world’s supercomputers in protein folding. But the success rate was very low until 2018. London-based developer DeepMind joined the competition with their neural networks, which mined gigantic datasets to create a new AI application called AlphaFold. AlphaFold calculates the most-likely distance between a protein’s bases and the angles of their chemical bonds to generate protein folding possibilities.

When AlphaFold joined the competition, a breath of fresh air was injected into the race. Of the 43 protein-folding problems given, AlphaFold got 25 correct. The second-place winner got three right. AlphaFold completely shattered the world of computational protein folding by predicting the ways various proteins fold based on their base sequences. Its impact on the pharmaceutical industry is priceless.

Delivering Drugs

Drugs can be delivered in a variety of ways: orally, rectally, through injection, and through IV. However, these methods all administer drugs almost immediately, affording patients and their providers little control over when the drug is delivered.

CRISPR is one promising technology that edits genes to deliver drugs. It can do many simple tasks in gene editing, and recently, researchers used CRISPR to create materials that shift their shape on command. This discovery could help doctors control exactly when and where the medicine should be delivered in the patient. In turn, drugs using the CRISPR method of delivery will be completely personalized to each patient’s lifestyle and disease severity.

Nanotechnology is another promising avenue for targeted drug delivery. Medical nano-robots have successfully been shown to fight cancer in humans. Other successful tests in nanotechnology surround the in-vivo operation of medical nano- and micro-robots.

The Future of Medicine

The COVID-19 pandemic is pushing the scientific and research community to cast aside national pride, research secrets, and publishing politics to share more information and discoveries. The hope is that these open lines of communication speed up our path to a cure. Ultimately, emerging technologies in Big Pharma and companies like Insilico are going to extend our lifespans and help prevent diseases.

Zhavoronkov places his prediction for extended lifespans at “maybe 20 years—that’s a reasonable horizon for tangible rejuvenational biotechnology.” What would you do if you found out in 20 years that you could extend your life by another 20 to 50 years? How would that change how you live today and your plans for tomorrow? Let us know your thoughts in the comments below!

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