Product1 publisher3 min readPublished
The six clocks agreeing matters more than the years they report, because a drug that quiets an inflamed lung moves the same blood proteins whether or not it slows ageing anywhere else.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Forty-two people with scarred lungs signed a second consent form and gave blood at baseline and at weeks two, four and twelve [3]. That is 59 percent of the trial's patients and the entire dataset behind the years [1].
Those samples were scored on 2,841 proteins against a reference population of 55,319 UK Biobank profiles [4]. Idiopathic pulmonary fibrosis is itself an inflammatory disease that shows up in blood proteins [8], which means a drug that quiets the lung and a drug that slows ageing can produce the same reading. The paper's authors say plainly that they cannot yet tell the two apart [7]. Their mitigation is that the 30mg twice-daily arm reversed age-associated protein patterns while the placebo group drifted along a normal ageing trajectory, which the analysis treats as suggestive rather than settled [9].
Michael Levitt, the 2013 chemistry Nobel laureate, said what convinces him is the agreement rather than the size of the effect, because the six models share neither their features nor their training data [11]. Independent models converging rules out one group's feature selection manufacturing the number by itself. It does not rule out a confound sitting upstream of all six, because all six are reading the same 42 people's blood.
The trial underneath was built to answer a different question. Its primary endpoint was safety, across 21 Chinese sites over 12 weeks, with adverse events at similar rates in every arm [14]. The 60mg once-daily group gained 98.4ml of forced vital capacity while placebo lost 20.3ml [15], a gap of 118.7ml [2]. Nintedanib and pirfenidone, the two approved antifibrotics, slow the decline in lung function rather than halt or reverse it [19], so a measured gain is the result worth having on its own terms. Alex Zhavoronkov, Insilico's founder and co-CEO, gets from the proteomic work to three years added to everyone's life and trillions of dollars in productivity and savings [10], which is a long way downstream of a week-four reading in one dose arm of a 42-person substudy [5].
A surrogate metric earns the right to drive anything if it predicts the outcome you care about in a population like the one you are measuring, and if it cannot also move for a reason your intervention already explains. Proteomic ageing scores clear the first bar: they predict mortality and the risk of common age-related disease across large and diverse populations, with organ-specific versions validated on similar cohorts [12]. They stumble on the second, because in patients whose disease is inflammatory and visible in blood proteins, the metric can move for exactly that reason [8]. A metric that predicts well and moves for reasons you already knew about is the one that makes a report look productive while teaching nobody anything, and the clocks sit in that quadrant here in a way they would not in a cohort with healthy lungs.
For anyone deciding what to do with this, the figure that survives scrutiny is the 118.7ml in a safety-primary study, and the biological age number stays a hypothesis with 42 people in it until a group outside Insilico reproduces it.
Ranked by verification strength, evidence, and original report placement.
Insilico Medicine published in Nature Biotechnology a secondary analysis of rentosertib, rather than a new trial.
Rentosertib is a TNIK inhibitor for idiopathic pulmonary fibrosis that Insilico Medicine designed using generative AI.
Of the 71 patients in rentosertib's phase 2a study, 42 consented to proteomic profiling, giving samples at baseline and at weeks two, four and twelve.
Some 2,841 proteins were measured on an Olink panel and scored against six clocks built by different groups using different methods, with 55,319 UK Biobank profiles as the reference population.
The largest effect appeared at week four in the 30mg twice-daily arm, where the chronological clocks put the treated patients between 2.7 and 3.5 years younger than expected.
Insilico's own summary of the work describes roughly three to four years, and up to six on certain clocks, which is the upper end of the range rather than its centre.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Everything traces to one paper
The specifics are unusually concrete -- 2,841 proteins, six clocks, 55,319 reference profiles, a 2.7 to 3.5 year window at week four -- but they all come from a paper Insilico wrote about its own drug, reported here by a single newsroom. The score holds up rather than down because the authors state the central limitation themselves, and because the proteomic data plus analysis pipelines are public, so the arithmetic can be redone by people with nothing riding on the answer. That recomputation has not been published yet, so the paper's numbers still stand unverified by anyone outside the sponsor.
Still a trial drug
Rentosertib remains a trial drug, treating patients rather than serving customers. The phase 2a work covered 71 people at 21 Chinese sites and the ageing analysis rests on the 42 who agreed to extra blood draws; phase 3 began in July and tests the lung. Since no regulator licenses a drug on a change in proteomic age, the clock result cannot advance a filing however well it holds up, which keeps real-world uptake at the level of one ongoing trial programme plus a public dataset.
Rounded up on the way out
The measured peak was 2.7 to 3.5 years, at one timepoint, in one dose arm. Insilico's summary of the same paper reports roughly three to four years and up to six on certain clocks, and its co-CEO converts three added years into trillions in productivity and savings. That is a considerable distance travelled from a 42-patient reanalysis whose authors say they cannot yet distinguish the drug's effect on an inflamed lung from an effect on ageing. The Next Web's own line — a hypothesis with a sample size of 42 — is the honest version.
The sponsor is also the author
The drug, the AI platform credited with designing it, the reanalysis and the quote about trillions all belong to one company, and a phase 3 trial started in July gives Insilico reason to keep the biological-age reading in circulation. Levitt's endorsement comes from outside and is careful about what it endorses, the clocks' mutual independence rather than the magnitude. The data deposit runs against interest, since it hands anyone who doubts the result the material to attack it.
One newsroom, caveats intact
The Next Web's account is precise and does not launder the weak points: it flags the seven hepatic discontinuations and their overlap with nintedanib, notes that Insilico quotes the upper end of its own range, and says the ageing figure is a hypothesis at n=42. It remains a single account of a single sponsor-authored paper, with no independent statistician, IPF clinician or regulator on the record, so our confidence attaches to what was reported rather than to what the drug does to ageing.
product
Insilico credits its AI with a drug discovery. The patent credits five humans.1 publisher
product
DESI ships 5.6 trillion pixels as a free viewer astronomers already open first1 publisher
science
Siderophore-engineered bacteria weather olivine 2.6 times faster in flowing seawater1 publisher
build
Vivodyne is spending venture money on wet-lab throughput, not bigger models2 publishers
Publishers with included, body-backed reporting in this cluster.
1 article · September 7, 2026