Abstract:
After taking a drug originally used to treat pulmonary fibrosis, the patient became 3 years younger after taking it for 4 weeks. This drug has been tied to AI from the beginning: AI first determines which target should be targeted in pulmonary fibrosis, and then generative AI designs the corresponding drug molecules.
The story is not far away from us. Insilico, which made it, has its main office in Hong Kong, and its important team responsible for new drug discovery and research and development has been based in Zhangjiang, Pudong, Shanghai for a long time. A few years ago, AI selected a target called TNIK from massive disease data. Now this judgment has been made into a real drug and has reached the second half of clinical trials.
In July this year, Insilico announced that rentosertib has officially entered Phase III clinical trials and plans to recruit approximately 320 pulmonary fibrosis patients in 47 centers.
Just as this lung drug continued to advance towards the end of the market, researchers found an unexpected signal from the blood samples left over from the previous Phase IIa.
42 patients, six different sets of "aging clocks" point in the same direction:
Their blood looks younger.
In this way, a lung medicine discovered by AI was suddenly related to "anti-aging".
01
The target chosen by AI a few years ago has reached Phase III
Before a drug can be officially researched, it must first find a biological target worthy of intervention. It may be a protein, a receptor, or a signaling pathway closely related to the disease process.
This object is the "target" of the drug.
The problem is that there are thousands of proteins in the human body, and diseases involve a large number of genes, cells and signaling pathways. In the past, finding new targets relied on long-term basic research, accumulating evidence layer by layer, and then slowly deciding which direction was worth betting on.
As the saying goes, draw the target first and then shoot the arrow. If there is a problem with the judgment of the target, the subsequent drug research and development will definitely go wrong.
AI can be said to have played a big role in finding targets.
It can read genomes, transcriptomes, proteomes, disease databases, papers and biological networks at the same time, and find which proteins and diseases have links worthy of further verification from information that is far beyond what humans can process one by one.
In recent years, AI participation in target discovery has become one of the important directions of AI pharmaceuticals. Nature Reviews Drug Discovery, the top review journal in the field of drug research and development, even published a review this year to discuss the involvement of AI in target identification, believing that it is playing an increasingly important role in this link. However, it also emphasized that whether a target is truly established ultimately depends on experimental and even clinical verification.
In this field, BenevolentAI and AstraZeneca have long used knowledge graphs and machine learning to screen new targets from complex disease data in chronic kidney disease, pulmonary fibrosis and other directions; Recursion has also been using large-scale phenotypic data and machine learning to find new disease mechanisms, such as its REC-4881. The key discovery behind it is that its platform has identified that "inhibiting MEK1/2 may treat familial adenomatous polyposis", and it has now advanced to Phase II.
It is not new for AI to help people find targets. The real difficulty is to advance the research.
The target selected by the algorithm must first be experimentally verified; after the verification is established, someone must design a truly usable molecule around it, and then go through toxicology, animal experiments, Phase I, and Phase II to prove that it is not just "theoretically reasonable."
Many targets discovered by AI will get stuck somewhere in the middle.
Phase I mainly looks at safety, and phase II begins with efficacy signals and doses; in phase III, hundreds or even thousands of patients are often recruited and compared with placebo or standard treatment under conditions closer to real clinical conditions. At this level, the previous answers that seemed "natural" to dozens of people often have to be re-verified. Seemingly obvious therapeutic effects may be diluted by larger samples, and adverse reactions not exposed in small samples may surface for the first time.
Because of this, Phase III is usually one of the most critical, most expensive, and most costly stages before a drug goes on the market.
The most noteworthy thing about Insilico this time is that it has pushed rentosertib into Phase III clinical trials. The company claims that this is the first case in which "AI discovers new targets + generative AI designs new molecules" enters the critical clinical stage.
This route can be pursued as early as around 2019. At that time, Insilicon was still using known targets such as DDR1 (discoid domain receptor 1) to prove whether generative AI "can design molecules." That year, they used AI to generate candidate molecules for DDR1 in 21 days and completed in vivo and in vitro verification, but it was more like a demonstration of ability. The questions were given by people, and AI was only responsible for solving them.
Insilicon is not content with finding molecules around known targets, but also uses their AI target discovery platform PandaOmics to find new treatment directions directly from disease data. Idiopathic pulmonary fibrosis became one testing ground, and AI eventually pushed TNIK (TRAF2 and NCK interacting kinase, a protein kinase involved in cell signaling) to the front.
Then, at the end of 2020, the candidate molecules designed around TNIK were officially identified as preclinical drug candidates, which means they have been screened out from a large number of candidate molecules and are ready to enter systemic animal experiments and subsequent clinical development.
It took less than 18 months from the start of target discovery to the creation of this drug candidate. Next, it’s the human body’s turn to answer whether this answer is reliable or not.
In February 2022, this drug, later named rentosertib, entered Phase I clinical trials; in July 2023, Phase IIa was officially launched in China, with 71 patients with idiopathic pulmonary fibrosis receiving 12 weeks of treatment.
In July this year, this drug officially launched Phase III large-scale clinical verification.
However, this is not the first drug involving AI to reach Phase III.
Less than a month ago, Moderna and Merck announced a case that was approaching the end of the line earlier.
On August 19, the two companies announced that the personalized mRNA cancer therapy intismeran autogene (V940/mRNA-4157) combined with Keytruda achieved positive results in the phase III melanoma trial INTerpath-001. The trial recruited 1,137 patients with stage IIB-IV melanoma who had been completely removed by surgery but still had a high risk of recurrence. The results showed that the combination therapy was better than Keytruda alone in the two key indicators of recurrence-free survival and distant metastasis-free survival. Moderna said that this is the first time that a personalized neoantigen therapy has received positive Phase III results.
The role that AI plays here is also very specific: Each cancer patient's tumor may contain a large number of different mutations. Moderna will sequence the patient's tumor and blood, and then use a set of AI algorithms to screen out from these mutations up to 34 "neoantigens" that are most likely to activate the immune system, and finally customize an mRNA therapy according to each person's different mutation combinations.
It is not the same AI pharmaceutical route as rentosertib. The problem faced by the former is "with so many mutations in a patient, which one should be selected to make a vaccine", while the latter is more similar to "with so many proteins and pathways behind a disease, which one is worthy of being developed as a new drug target."
But when the two projects are put together, we can see that the answers given by AI participation are moving out of computers and early experiments and into Phase III clinical trials that truly determine the fate of the drug.
02
A lung medicine that makes blood "look younger"
Just as rentosertib was preparing to undergo the Phase III test, researchers found an unexpected signal from the blood samples left in the previous Phase IIa:
Forty-two patients, six different sets of "aging clocks", finally captured the signal of moving in the "younger" direction.
The most consistent of these was the 30 mg twice daily dose group, with five of six sets of clocks pointing to "younger" at week 4. In the once-a-day 60 mg group, the four sets of clocks based on actual age training gave a predicted age decrease of about 2.7 to 3.5 years.
In other words, the aforementioned "eating for 4 weeks makes you 3 years younger" is not entirely a metaphor - at least judging from proteomic indicators, "rejuvenation" is indeed possible.
Youth here mainly depends on the overall state of thousands of proteins in the blood.
The research team re-analyzed the blood samples left from the Phase IIa clinical trial. Forty-two patients were ultimately included in the proteomic analysis, and they had blood samples before taking the drug, as well as at weeks 2, 4, and 12.
The researchers measured thousands of these proteins at once and put the data into six independently developed "proteomic aging clocks."
The so-called aging clock can be understood as an age prediction model. As people age, many proteins in the blood will show relatively stable changes.
It's a bit like "guessing your age by looking at your face" - people of different ages often show some regular changes in wrinkles, skin condition, and facial structure. Once the model has seen enough data about people of different ages, it can estimate how old a person roughly looks based on these characteristics.
The proteomic aging clock does something similar, except that instead of looking at faces, it looks at combinations of hundreds or thousands of proteins in the blood.
So what the researchers are really asking is: After taking rentosertib, does the blood of these patients become more like young people or more like old people at the protein level?
The answer, at least in Week 4, very consistently points to the former.
At this point, I have to mention that this problem was not actually something the researchers thought of after the fact.
On the one hand, idiopathic pulmonary fibrosis itself is a highly age-related disease, and the reason why TNIK was initially targeted by Insilico is not just because it is related to fibrosis. In early analysis, TNIK was also linked to multiple classic aging mechanisms.
On the other hand, Insilicon has been conducting research on aging and longevity for a long time, so during the Phase IIa clinical trial, the research team left continuous blood samples in advance, hoping to see whether a drug originally used to treat lungs would encounter the issue of "anti-aging".
Interestingly, the "youth" signal found this time and the improvement in lung function do not completely overlap.
In the previous phase IIa, the most obvious improvement in lung function was in the 60 mg dose group once a day; but the most stable change in the aging clock was in the 30 mg twice a day group. The researchers then compared changes in FVC, or forced vital capacity, with changes in predicted biological age, and found that there was no strong correlation between the two.
This means that the blood looks younger, at least it cannot be simply explained as "just because the lung disease is better."
The research team also compared the protein changes in these patients with the natural aging trajectory of more than 50,000 elderly people in the British Biobank. In the 30 mg twice daily group, some of the protein changes were indeed opposite to normal aging.
A drug originally used to treat pulmonary fibrosis has a completely different hypothesis:
Will it change more than just the lungs?
03
"3 years younger", far from the real
How far is anti-aging?
As mentioned before, Insilicon has been conducting research on aging and longevity for a long time.
This is inseparable from the founder Alex Zhavoronkov, who himself is a fairly typical "longevityist" in this field. Zhavoronkov left the IT industry when he was about 24 years old and turned his head into aging research. The reason is very simple - if humans really want to one day go further into the universe, we ourselves must first live longer.
He has been walking this road for more than 20 years.
In recent statements, he has repeatedly emphasized that humans can already enter space, communicate with machines, and create virtual worlds, but it is still very difficult for them to live a few more years.
Faced with this "3 years younger" result, Zhavoronkov seemed quite restrained.
In an interview with the Wall Street Journal, Zhavoronkov directly poured cold water on the statement that "AI will soon double human lifespan." He believes that no drug has yet undergone rigorous human clinical trials and has been proven to truly extend human life.
However, he also said that if the goal is lowered from "doubling life span" to "reversing part of biological age", he believes that it is not completely impossible in the next ten years.
And the question that this paper really needs to answer is here:
What does the six sets of aging clocks going back together mean?
First, the sample is small. This proteomics analysis ended up with only 42 pulmonary fibrosis patients. After being divided into four groups, there were only about 9 to 11 people in each group. Although the six sets of clocks have the same direction, they still analyze the same batch of patients and the same set of proteomic data.
Six models yielding similar answers is certainly more convincing than just picking the best-looking result, but it does not mean that six independent experiments have replicated the result.
The more critical issue is that pulmonary fibrosis itself changes a large number of proteins related to inflammation, metabolism and tissue repair.
The "youth" that researchers see, is the drug really touching the core of aging, or is it just that after the disease improves, the blood naturally becomes closer to the state of healthy young people? This question is also one of the most asked questions during peer review of papers.
Although the authors have presented some evidence that the "younger" signal and improved lung function are not entirely synchronized, this is not enough to rule out another, more common explanation. After all, FVC can only reflect part of lung function. Pulmonary fibrosis also affects a large number of systems such as inflammation, metabolism, and tissue repair, and these changes will also be reflected in blood proteins.
The paper ultimately admitted that the existing data cannot truly separate the "anti-aging effect" from the "associated changes caused by disease improvement."
There is also a less beautiful, but very important detail: the most obvious "younger" signal is concentrated in the 4th week, but it does not go all the way down as you continue to take the medicine.
By week 12, some of the effects of the aging clock have diminished. During the peer review, the reviewers also specifically asked: If the drug is really continuing to change the aging process, why do the most beautiful results appear at an intermediate time point?
So, this study is now more appropriately called an exploratory anti-aging signal, which is far from "proven that anti-aging is effective."
Vadim Gladyshev, a Harvard aging expert interviewed by the New York Times, also used a similar scale. He believes that this is a very clear signal of "predicting biological age decline", but he also emphasizes that the sample is small, the aging clock itself still has limitations, and this drug has not yet verified this effect in healthy people.
AI certainly hasn’t found the elixir of life yet.
But this still leaves a lot of room for imagination. The most interesting thing about rentosertib is not that it temporarily moves a number back three years, but that it strings together several things that were originally far apart:
AI first proposes a target from complex biological data, and generative AI then designs molecules around this target. After the drug enters the human body, researchers begin to ask in turn: In addition to treating a specific disease, will it also change some underlying aging processes?
We can believe that AI is turning some biological conjectures that were extremely difficult to screen and verify in humans in the past into real drug experiments more quickly.
As for whether there is really a medicine in these experiments that can finally turn back the human clock - at least now, this question is no longer so far-fetched.
For rentosertib, the next step is actually very simple: first look at whether the target selected by AI in Phase III a few years ago can really improve more patients with pulmonary fibrosis.
Let’s get this over with first, and then talk about what is hidden behind “3 years younger” will have more weight.
Comments