Here’s the ongoing revolution of artificial intelligence in medicine production. Economist Report
Artificial intelligence in drug discovery and testing: GSK, Insilico Medicine, digital twins, and new biological models. The Economist article.
Patrick Schwab is no ordinary pharmaceutical researcher, and his workplace is no ordinary pharmaceutical laboratory. There are no benches or bubbling liquids. Even white lab coats are missing. Dr. Schwab, instead, is dressed entirely in black. But this is the appropriate attire for someone who works in King's Cross, an area once occupied by railway yards and industrial buildings, but now, after a transformation, is one of London's most fashionable neighborhoods, writes The Economist .
Dr. Schwab works for GSK, a pharmaceutical company. His job is to rethink the future of drug manufacturing using an equally trendy branch of computer science, artificial intelligence (AI). He's applying this technology to shift as much of the workload as possible from the glassware to the computer: designing drugs in silico, rather than in vitro.
To this end, he is developing a software tool called Phenformer, which he is training to read genomes. By linking genomic information with phenotypes, the biological term for the physical and behavioral outcomes of particular genetic combinations, Phenformer learns how genes determine disease. This allows him to generate new hypotheses about diseases and their underlying mechanisms.
INSILICO MEDICINE AND THE TRANSFORMERS
Insilico Medicine, a Boston-based biotech company, appears to have been the first to apply the next generation of AI, known as transformer models, to drug discovery. In 2019, its researchers wondered whether they could be used to invent new drugs based on biological and chemical data. Their first target was idiopathic pulmonary fibrosis, a lung disease.
They began by training an AI on datasets related to this condition and found a promising protein. A second AI then suggested molecules that would bind to that protein and modify its behavior, but that weren't too toxic or unstable. Human chemists then took over, creating and testing the selected molecules. The result was named rentosertib and recently successfully completed the mid-stage clinical trials. The company says it took 18 months to develop a candidate, compared to a more typical timeframe of four and a half years.
Insilico now has a pipeline of over 40 AI-driven drugs for the treatment of conditions such as cancer, intestinal diseases, and kidney disease. And its approach is gaining traction. Annual investments in this area are projected to rise from $3.8 billion to $15.2 billion between 2025 and 2030. […]
DRUG ECONOMICS AND THE BENEFITS OF AI
Given the strange economics of the pharmaceutical industry—drugs entering clinical trials have a 90% failure rate, pushing the cost of developing a successful drug to as much as $2.8 billion—even marginal improvements in efficiency would offer major benefits. Reports from across the industry suggest that artificial intelligence has begun to deliver these results. It has shortened the preclinical phase (before the start of human trials) from three to five years to 12 to 18 months. Furthermore, it has improved the success rate. A study published in 2024 on the performance of molecules discovered by AI in the early stages of clinical trials found a success rate of 80 to 90%. This compares with historical averages of 40 to 65%. […]
AGENTS AND EXPERIMENTAL DESIGN
Artificial intelligence is also helping improve trial design. One approach involves using AI "agents" that act as if they can think and reason. Back at GSK, Kim Branson, head of AI, showed your correspondent a demonstration of an agent-based system called Cogito Forge. When presented with a biology question, Cogito Forge can write its own code to help answer the question, collect the appropriate datasets, merge them, and then create a presentation complete with graphs showing its conclusions.
From there, it can generate a hypothesis about a disease, including testable predictions, and attempt to verify or disprove it with a literature search. This search employs three agents: one to search for reasons why the hypothesis is valid, a second to search for reasons why it is not, and a third to judge which of the other two is correct.
PATIENTS, DATA, AND DIGITAL TWINS
Another area where AI shows promise is patient selection for trials. It can analyze candidates' medical records, biopsies, and body scans to identify those who would benefit most from a new drug. Better participant selection means smaller, and therefore faster and more cost-effective, trials.
The most intriguing use of AI to improve trials, however, is the creation of synthetic patients (sometimes called digital twins) who serve as controls matched to real participants. To do this, an AI examines data from past trials and learns to predict what might happen to a participant if they followed the natural course of their condition rather than being treated […]
UNLEARN.AI AND SIMULATIONS
If adopted, the use of synthetic patients would reduce the size of trial control arms and potentially eliminate them altogether in some cases. They could also be attractive to participants, as they would increase their chances of receiving the trial treatment rather than being placed in a control group.
A model published in 2025 by Unlearn.AI, a San Francisco-based digital twin company, suggested that this approach could reduce the size of a control arm in an early-stage Parkinson's disease clinical trial by 38% and by 23% in another Alzheimer's disease study. Furthermore, early-stage clinical trials, which sometimes lack a control arm, could now introduce them digitally to increase confidence in early signs of efficacy and improve the design of subsequent studies.
BIOLOGICAL STRUCTURES AND VIRTUAL MODELS
Many proteins, molecules increasingly used as drugs but much larger than conventional drug molecules, tend to move. This makes it more difficult to determine their precise shape. RNA molecules, the basis of a new class of vaccines, are equally complex. And the complex membrane structures found inside cells are even more so. But this is an area where understanding is rapidly advancing. AIs are now being trained to model the interactions between proteins and other molecules, predict the folding of RNA, and even simulate virtual cells. […]
INDUSTRIAL BALANCE AND COLLABORATIONS
These departures from standard drug discovery tools raise the question of whether conventional pharmaceutical companies are at risk of disruption. OpenAI, in particular, has made clear its expectation that models will reach high levels of capability in biology and is training systems capable of reasoning and making discoveries in the life sciences. For now, pharmaceutical companies have the advantage of having a wealth of biological data and the context to understand and use it. Collaboration is currently the order of the day. OpenAI, for example, is working with Moderna, a pioneer in RNA vaccines, to accelerate the development of personalized cancer vaccines. But this balance of advantages could change.
EFFICIENCY AND IMPACT PROSPECTS
Whoever gains the upper hand, however, if AI can achieve similar efficiencies from clinical trials, the probability of a molecule successfully completing clinical trials could increase from 5-10% to 9-18%. This may not seem like much, but it represents a huge reduction in risk for the company, resulting in lower drug development costs. In the medium term, this could stimulate investment and increase the number of drugs brought to market. In the long term, if AI can solve biological problems, the technical possibilities for improving human health could be almost limitless.
(Excerpt from the foreign press review by eprcomunicazione )
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/sanita/ecco-la-rivoluzione-in-atto-intelligenza-artificiale-nella-produzione-di-medicine-report-economist/ on Sat, 10 Jan 2026 05:43:27 +0000.