Drugs designed by artificial intelligence have never participated in phase 3 clinical trials. This is the final hurdle before regulatory approval. However, this can change soon. Small molecules in in silico medicine that target chronic lung diseases can enter this final stage within next year or two.
The company reported a study of 71 patients in China in June, indicating that the molecule, Rentseltib, is safe and well tolerated. It is the most advanced research drug that biological targets and therapeutic compounds have been discovered using AI.
Nowadays, there are periodic claims that AI is set up to “translate drug discoveries.” However, there are holdouts warning that AI has not yet had a major impact at the critical and costly stages of drug research and development. Insilico should demonstrate that its candidates effectively treat idiopathic pulmonary fibrosis, a condition that covers the lungs in a large phase 3 trial.
Discovering turbo charging
AI is very efficient at searching and mining huge datasets, summarizing that information and detecting patterns of data. “The promise of AI is to be faster and a little more sensitive when detecting signals in a sea of loud noise,” says Chris Meyer, a former drug researcher at the Boston Consulting Group (BCG). This is especially true for machine learning, for example, which is a type of AI that is very skilled to become very skilled in which proteins or genes to target based on experimental results.
The promise of AI is to be faster and a little more sensitive when detecting signals in loud oceans
Kris Meyer, Boston Consulting Group
Meier led the first systematic analysis in 2022 to evaluate the clinical pipeline of AI-driven biotech companies. This is because the molecules discovered by Phase 1 AIs achieved success rates of 80-90%, with a historical average significantly higher than about 66%. AI can revolutionize biological discovery, analysis explained, and conducted the most time-consuming, repeated, and expensive procedures and turbocharge exploration.
Not everyone was impressed. AI relies on collected data. In other words, the chemical space has already been investigated. “Many of these targets are less novel,” says Andreas Bender, a professor of machine learning in medicine at Khalifa University in Abu Dhabi. “You know the target, so you know more about the safety associated with it.” This will help businesses select the right candidate to pass the initial safety trial.
Drug chemist Derek Lowe scrutinised each of the 24 candidates on the blog, noting that in almost every case, “the target is already known to be involved in the disease being investigated.” He is wary of AI being oversold in a “barbuilt of stories about extraordinary progress,” and there are revolutionary claims. “People get hooked in this area very easily,” Lowe says. “I've seen the waves after the wave of enthusiasm for all sorts of different approaches to calculation.”
AI is good at predicting protein structures, but here an algorithm like Alphafold sits in a large, clean dataset from Protein Data Bank, which contains over 200,000 structures. When using scientific literature for machine learning, algorithms encounter problems rather than problems with their creation. “To get traction, machine learning models need a lot of negative consequences of the quality scattered across the data. But we don't publish them — we pretend they never happened,” says Lowe.
The big thing
Still, the partnership between AI Biotechs and the Pharma company is growing. In January 2024, Isomorphic Labs, a subsidiary of Tech Giant Alphabet, entered into a deal with Novartis with pharmaceutical company Eli Lilly, which could be worth billions. In 2023, Benevolent signed a contract with Merck for $594 million (£439 million).
Recursion was founded in 2013 on a machine learning strategy based on a single basic assay and now runs a highly automated in-house lab. “We generate data covering biology and chemistry swaths before focusing on individual diseases,” says Lina Nilsson, the company's chief platform officer. The company's stated purpose is to reduce the massive 90% failure rate of traditional drug discovery.
One highlight from the pipeline is molecular adhesives that lead to the degradation of proteins essential for the survival of some cancer cells. Recursion announced last December that it had administered the first patient with the candidate drug to treat certain solid tumors and lymphomas. 'This program has arrived from target initiation [new drug application]- We do the research in about 18 months, with an industry average of around 42 months,” says Nilsson.
However, Lowe says that AI is best equipped to support in areas “which are almost the opposite of how important it is to drug development costs.” He points out that he is good at proposing new molecules that target known proteins or genes, but how to make them, but he does not sweat much against patient pitfalls such as unexpected toxicity.
Later fails cost more. If the compound appears promising in the lab, we will go to a Phase 1 clinical trial to assess safety. Such trials involve only 15-20 people and cost millions of dollars. The next phase will cost around $45 million, and the final phase 3 study on effectiveness takes three to four years, allowing hundreds of millions of dollars to be swallowed, and a 55% chance of success.
Hype vs. hope
This is related to AI drug candidates because they are not always dismantled at the biological target level, Lowe says. '[InSilico] We tried to speed up the early stages of the process using machine learning technology, but then they get up and take the opportunity just like everyone else in us,” he says. He is not alone in this view.
Benders say AI thrives by analyzing the results of biochemical assays to select targets and hits, which is important, but “many of the in vivo translation is completely missing.” Why: There is much less data from patients, and for now it is a low-performance toxicity model. “There are millions of data points for target-related activities, but there are far fewer data points for complex biological systems,” says the vendor. “In many cases, we cannot extrapolate from simple assays to prediction of organ toxicity,” he adds. Even the mechanisms of drug-induced liver damage, the most common type of organ toxicity, are only partially understood.
I am a short-term pessimist and a long-term optimist. I don't think there's a reason why these techniques can't do great things
Derek Lowe, drug chemist
For now, all of this leads to conclude that there is a gap between hope and reality, but it doesn't have to stay that way. “There is some evidence that AI accelerates discoveries and makes them more successful in some places,” says Meier of BCG. “The hope is that AI will make drug discovery faster and change its economy faster.” However, many drug candidates who are heavily affected by AI have not reached a large patient trial to test their efficacy, making it difficult to communicate right now.
Last year, AI-focused biotechnology hit headwinds. AI-Drug company Exscientia has cut a significant staff cut in 2024 and narrowed its pipeline. Germany-based biotech Evotec has trimmed its pipeline 30% this year. And in May, Recursion announced that it had ended two clinical programs for rare neuropathy and one clinical program for treatment Clostridium difficile Infected for commercial reasons. “We prioritized oncology and rare diseases,” says Nilsson.
Drug candidates developed using AI are now mainstream, but it remains unclear how successful the current generation will be. Nevertheless, experimental automation, as well as machine learning and generative models, are prevalent in drug research and development. AI is ultimately the view of skeptics. “It's not a good idea to oppose progress driven by the capabilities of computing, hardware and algorithms,” Lowe says. “I'm a short-term pessimist and a long-term optimist. I don't think there's a reason why these techniques can't do great things.”
