Our mission to make business better is fueled by readers like you. To enjoy unlimited access to our journalism, subscribe today.
Artificial intelligence is making rapid inroads in helping with the discovery of potential medicines.
In the latest example, a Hong Kong–based biotechnology company, Insilico Medicine, which uses A.I. tools to help it find potential new therapies, announced Tuesday it has brought a drug candidate from an initial scientific hunch to the cusp of human clinical trials in less than 18 months, a time span the company says may be a new record for a process that often takes more than four years.
The company said it was the first time A.I. had been used to find a completely new target pathway for a disease and then generate a new molecule to hit that target, and to do so in such a shortened time frame. Alexander Zhavoronkov, Insilico’s chief executive, describes A.I. as “a Ferrari” that allows the company to move quickly through the first few phases of early stage drug discovery.
The drug Insilico is developing is a potential therapy for idiopathic pulmonary fibrosis (IPF), a rare condition that affects older adults in which the lungs become scarred for unknown reasons, making breathing increasingly difficult.
Bringing a new drug to market is risky, time consuming, and expensive. On average, it takes a decade and costs more than $1 billion. Most new drug candidates fail—in fact, one recent study published in the Journal of the American Medical Association estimated that only 14% of drugs that enter initial human clinical trials ultimately make it to market. But many more fail before they even reach that stage.
A lot of the cost and time in bringing new medicines to market is tied up in the early drug discovery and preclinical phases, before any testing on human subjects begins. It’s estimated that between 45% and 75% of the cost of bringing a new drug to market comes in these early stages.
A number of companies worldwide are using artificial intelligence techniques to try to improve the odds, shorten the time, and lower the cost of early stage drug discovery. “Generating a target hypothesis is a big part of drug discovery, and it is the one with the largest failure rate,” Zhavoronkov says.
With its A.I. technology, Insilico spent less than $2.7 million to go from nothing to bringing it through all preclinical validation stages, he says, a fraction of what that process normally costs.
To find a target that may play a large role in IPF, Insilico used a software platform it developed called PandaOmics that incorporates several machine-learning methods. It examines a large database of RNA and proteins to identify those likely involved in the scarring process seen in IPF. It then identifies the genes most likely involved in regulating these processes and then filters them to find those with characteristics that may make them good drug targets.
To help choose among these potential targets, Insilico’s PandaOmics uses an A.I. module that can process language. It scanned medical literature, research papers, clinical trial data, and patent and grant applications to see if the particular target had been implicated in processes that may be related to IPF. “It is looking at text and evaluating how much additional evidence has been published around this target but necessarily about it in the context of this specific disease,” Zhavoronkov says.
To have a successful drug discovery target, he says, it’s important to trade off between confidence that your target will have an impact on the disease and novelty. If the target is too obvious, it almost certainly will have been tried before by other researchers, and there may be existing drugs that act on it, or there may already be evidence proving the target is ineffective. “Our target is novel in that it has never been in clinical trials and never mentioned in the context of fibrosis,” Zhavoronkov says.
Insilico also has an A.I. module that tries to predict the outcomes of Phase II clinical trials, the stage at which most new drug candidates that make it to human testing fail, and it uses this to help further narrow its targets.
Insilico ultimately selected 20 targets for experimental validation. Five of these worked well, with one target, which is implicated in fibrosis in several different organs, looking particularly promising. Once it found the target protein, Insilico had to find the crystal structure of that protein and then find molecules that may act on it.
To do this, Insilico used an A.I.-based chemistry software platform, called Chemistry42, that the company has licensed to a number of large pharmaceutical firms, including Merck, the German drugmaker. This enabled it to generate a ranked list of 80 likely molecules.
It then manufactured nearly 50 of these molecules and ran experiments to find the most suitable candidate. Zhavoronkov says this is between one-fifth and 1/10th the number of molecules that even other A.I.-powered biotech firms have had to synthesize in order to find a good potential drug for a new target, and about 1/50th the number that traditional pharmaceutical companies may have to try before hitting on the right candidate.
Tests of this molecule in mice showed a significant improvement in lung function compared with the existing treatment, a drug called nintedanib, made by drug company Boehringer Ingelheim. It also was found to be more potent in lab tests of fibrosis-related processes, including in donated human lung tissue. Now the company is hoping to put this molecule into human clinical testing, hopefully before the end of the year, Zhavoronkov says. If the tests are successful, a drug might make it to market within a few years.
Insilico, which Zhavoronkov founded in 2014, has received about $52 million in funding, including from Baidu Ventures, the venture capital arm of the Chinese search giant, and the Longevity Vision Fund, a venture fund dedicated to companies interested in extending human life spans. To date, Insilico has made money licensing its three software platforms— PandaOmics, Chemistry42, and InClinico, its Phase II clinical trial success algorithm—to large pharmaceutical and chemical firms and to hedge funds that want to make trades based in part on forecasting clinical trial news.
Insilico recently announced a partnership with Syngenta, the Swiss agricultural chemical business, to work on more innovative ways to protect crops from disease and weeds.
It has previously used its software to help find molecules that could potentially be used as drugs to help treat COVID-19.
