Every year, there are thousands of new drugs in development at labs around the world. But only a tiny fraction of them make it through to human trials -- never mind final approval. And even among those that do get through, the majority of new medications are actually just new-er versions of existing drugs, like cheaper generic versions. But there’s a new hero in town with the potential to help things along: drug discovery systems based on artificial intelligence. Even though the tech is still growing, several new drugs designed with the aid of AI are already in clinical trials. And together with other modern tech like fully automated robot scientists, AI is changing drug discovery in revolutionary ways -- and it’s accelerating the process beyond anything we’ve seen before.
It’s only recently that we’ve been able to create drugs from scratch. Some drugs are based on traditional remedies and natural products -- like aspirin. The first fully synthetic drug, the sedative chloral hydrate, was developed in 1869. But there was still quite a bit of trial and error involved. Rather than throwing spaghetti at the wall to see what sticks, it was… throwing chemicals at people to see if they stop being sick. That’s how, in 1932, we got sulfa drugs, when researchers at a chemical dye company discovered some of those dyes could be used to kill dangerous microbes. And pure random luck also did a lot for the early pharmaceutical industry. Most famously, if Alexander Fleming had been a little more fastidious about keeping his lab clean, we might not have penicillin.
But all that has changed. Starting in the latter half of the 20th century, we’ve seen the rise of rational drug design. That means no more spaghetti-throwing. Instead, scientists build drugs from the ground up, driven by hypotheses of how they might work. Having so much data to rely on can be a double-edged sword, though. In fact, in a sense, you could say we’ve gotten too good at this. The number of researchers and person-hours it takes to investigate every single drug candidate is truly monumental. But now, artificial intelligence is turning the tables by helping to guide and accelerate this process. It’s helping identify drug candidates we might not have found otherwise, and cutting years off their development. AI systems can be given a general description of what we want to find, analyze a bunch of literature and databases, and pick out the best hits for later research.
This starts right at the beginning: before you even think about how your new drug would work, you have to decide what to design it for. You need a target. That target is one of the dominoes in the sequence of things that happen in your body to create a disease. It could be a mutated gene, or an enzyme that’s working harder than it should. There’s usually a huge body of scientific literature devoted to the disease you want to treat -- stuff like research articles, clinical trial reports, and patient records. Deciding on a target based on all this information is like looking for a needle in a haystack, and it’s a lot to ask of our brains. AI can help us see the big picture -- then pick out the one really specific thing we needed from that big picture.
AI systems use a bunch of different technologies to do that, but generally, the key is natural language processing. Natural language processing is basically what allows your phone’s voice assistant to understand you when you say “Play something by Tchaikovsky.” If you’re lucky, it will understand that you want to hear any music composed by Tchaikovsky, instead of a specific piece he wrote called “Something.” Now, the AI doesn’t really understand what a “Tchaikovsky” is. But by analyzing the relationships between the words and the context, it can tell that you want to look for “Tchaikovsky” in the part of its database that holds a list of composers. And the principle is the same with drug discovery.
So even though the AI doesn’t understand what a “gene” is, if you instruct it to sort through the literature on a given disease, it can identify the ones that stand out. It can also find the places in the literature that talk about that gene -- and look at the words connected with it. Do authors refer to it often in the context of that disease? And crucially, do they talk about a causal relationship? If so, the AI can conclude that that gene is worth a look. Natural language processing also means that an AI system doesn’t need this info to be pre-formatted by an army of people. You can feed it things written by a human, like scientific articles or medical case reports.
So AI-based systems can analyze enormous amounts of data and give us meaningful answers. And they’re much faster readers than we are, so they can go through way more information, and handle way more complexity, than humans alone. Plus, they never need to duck out for coffee. This vastly speeds up the process of finding potential drug targets that we might have found eventually, but it also makes it possible to find completely new targets that we might not have noticed.
For example, researchers at the US company Berg grew cancerous and healthy cells from over a thousand donors in petri dishes. They varied the conditions and tracked the unbelievably complex chemistry that results from cells simply doing their thing. They ended up with trillions of pieces of data from those samples. That’s a lot of raw info about what happens in a cancer cell compared to a healthy one -- which is to say, too much. The amount of data they had was so gigantic that without AI, it would have effectively been useless. But their AI system was able to analyze this data and identify various molecules that were out of whack in the cancer cells compared to the healthy ones. This suggested a new cancer drug target – a molecule called coenzyme Q10. And a new drug candidate they developed based on that AI discovery is now in clinical trials for pancreatic cancer and squamous cell carcinoma.
So now we know what we want to target with a novel drug. Which means we need… a novel drug. Meaning it’s time to identify and synthesize chemical compounds that will hopefully hit that target in just the way we want. But how do you choose what to synthesize? Even before AI helpers, researchers were able to make educated guesses about what would work, by looking at the chemical features of substances that they already knew could interact with the target. That interaction doesn’t have to be beneficial – if something sticks to your target, that gives us a clue for how to design something new that will stick to it as well. But still, even if researchers know what chemical features they want their drug candidate to have, that may still mean thousands of options. And AI systems come to the rescue here as well. Scientists can tell the AI what chemical parameters they’re looking for, and the system will not only find suitable candidates, but cut the list down to those that will potentially work best.
The process is kind of similar to things like using AI for image recognition. When you do an image search for “cats,” you may get 99 images of felines, and one gorgeous kitten-looking cloud. Because of those occasional blunders, the AI doesn’t actually make the decisions. Scientists still evaluate the results, and the AI system only helps automate and accelerate the grunt work. But, like… that’s really helpful.
For example, by using AI to help choose chemical compounds, the makers of a new candidate drug for obsessive-compulsive disorder were able to cut down their development cycle from around five years to one, and get their candidate into clinical trials. Once you’ve decided on a drug candidate and synthesized the right chemical, it’s time for testing. But before human tests, and before animal tests, researchers start with simpler assays. That means testing the chemical using cultured cells or cell-free cocktails containing the drug’s potential target. It used to take years to do this preliminary testing for a single potential medication. But in the 21st century, pharmaceutical companies have turned to robotic high-throughput screening, which makes it possible to test hundreds of thousands of compounds in a single day. A human might have to pipet the hundreds or thousands of candidate compounds into one cell culture dish at a time, but a robot can quickly zip through a bunch of them.
A researcher can just design the experiment and then check on the results. Like the Voltron of modern pharmaceutical science, a bunch of autonomous capabilities can also be combined into what’s called a robot scientist. That system uses AI to identify specific experiments with a lot of potential, and can then autonomously use lab equipment to perform these experiments and fine-tune its decisions of where to go next based on their results.
For example, Eve, an AI-equipped robot scientist at the University of Cambridge, has already identified a new potential treatment for malaria. Eve first identified a list of compounds that may counteract malaria, then screened them against yeast cells in culture to see which of those chemicals worked best. This way, the Cambridge team was able to report that Eve had identified the well-known antimicrobial compound triclosan as a candidate to help combat treatment-resistant strains of malaria.
Even with all of this futuristic tech, though, only a few dozen new drugs are approved every year. For example, in 2019, the US Food and Drug Administration approved 48, but only 20 of them were actually distinct enough from existing medications to be considered truly new. That’s because discovering a new drug and bringing it to market is an expensive, long and difficult process. To develop one new drug candidate, researchers need to screen up to 10,000 compounds. And on average, only five of those turn out to be good enough to go into clinical testing. If it gets that far, a drug candidate needs to complete three phases of clinical trials. Of those, 90% fail to gain FDA approval. And that’s why developing a new drug may take up to 15 years, and cost around 1.3 billion dollars. Each.
But smart AI systems like the ones we’ve talked about are in a position to change that. We’re only beginning to see the results of this AI revolution in drug discovery. There are literally hundreds of companies developing AI-based systems for the pharmaceutical industry. Human scientists still drive the process. And while machines are great at finding patterns and sorting through a lot of information in a hurry, our brains can do a whole lot that machines can’t. But the help that AI represents is already truly revolutionary. It can cut the time and cost necessary to develop a drug. And by doing so, it’s widening that pipeline of potentially life-saving treatments. When candidate drugs inevitably flunk out -- because they don’t work, or aren’t safe for us to use -- this technology ensures there will be more waiting for us to test. And that’s great for all of us -- even if it means the future looks kind of like a robot handing us a miracle pill.