Seventy-one patients is a rather small audience for an idea with such large ambitions. In a Chinese trial of rentosertib, people with idiopathic pulmonary fibrosis received either an experimental drug or a placebo for twelve weeks. Their lungs, scarred by disease, became the testing ground for a proposition that had begun in a computer: could artificial intelligence help choose the biological target, then design a molecule worth giving to a human?
Insilico Medicine had done both. The trial, published in Nature Medicine in 2025, was designed primarily to assess safety. It also produced an encouraging lung-function signal. Some participants discontinued treatment; liver toxicity and diarrhea were among the events leading to discontinuation. It is the sort of result that demands another experiment. A lung has no interest in how elegantly its medicine was designed.
- The work: AI-assisted target discovery and molecular design, followed by laboratory and clinical testing.
- The customers: pharmaceutical R&D teams and researchers buying tools, collaborations or drug rights.
- The economics: software licenses plus upfront payments, milestones and possible royalties.
- The next test: a planned 320-person Phase III study of rentosertib, treating patients for 52 weeks.
The lung is the judge
Fibrosis is what happens when the body's repair work becomes destructive. Scar tissue accumulates. In the lungs, the consequence is progressively impaired breathing. Insilico's candidate inhibits a protein called TNIK. The company's distinction lies in the sequence: its software helped prioritize TNIK as a fibrosis target and then helped produce a molecule aimed at it.
The sequence matters because finding a molecule and choosing the right disease mechanism are different problems. A beautiful key is useless when you have chosen the wrong lock. Insilico wants to improve both decisions. The human trials examine whether those decisions survive contact with a disease that is considerably less tidy than a model.
An aging question becomes a chemistry problem
Founded in 2014 by Alex Zhavoronkov, with scientific co-founder Alex Aliper, Insilico grew out of an interest in aging and computation. Zhavoronkov had worked at graphics-chip company ATI Technologies before building a biotechnology business. The path from graphics processors to lung disease is unusual, but it helps explain the company's instinct: look for a biological problem that might yield to computational scale.
In 2019, the team published a timed experiment using its GENTRL model to design DDR1 inhibitors. Six compounds were designed in 21 days; four were active in biochemical assays and two validated in cell-based assays. One was tested for pharmacokinetics in mice. The exercise established that generated structures could become experimentally interesting chemicals. It did not establish that a medicine could be delivered in three weeks.
Over the following years, the business acquired the apparatus of a drug developer. PandaOmics and Chemistry42 launched in 2020. A $255 million Series C followed in 2021. Feng Ren, who had spent eleven years at GSK, joined as chief scientific officer that year. His background supplied a discipline that computation alone could not: turning promising chemistry into a development program.
Three tools, one expensive question
Pharma.AI divides the work into useful questions. PandaOmics asks which biological targets deserve attention, combining molecular datasets, networks and published research. Chemistry42 generates and ranks molecular structures against chosen objectives. inClinico forecasts clinical outcomes from several kinds of data. Together they attempt to shorten the route from a disease hypothesis to a candidate and a development decision.
inClinico adds forecasts. Experiments supply the evidence.
The customer is usually a pharmaceutical scientist, biotechnology team or research institution. A medicinal chemist can explore candidate structures; a biology team can examine target hypotheses. A pharmaceutical partner can instead commission discovery work or license an asset. Insilico sells several entry points into the same process, rather than requiring every buyer to purchase the whole machine.
Its competition includes Recursion, which also connects computational biology, chemistry and clinical development. Conventional discovery teams remain an alternative. Insilico's distinctive proposition is the combination of commercially available tools and an internal pipeline that puts those tools to work. Its own candidates make it a demanding customer of its own software.
The $2.6 million footnote
The company's account puts the fibrosis program's hypothesis-to-preclinical-candidate work at approximately $2.6 million and just under eighteen months. A peer-reviewed report describes roughly eighteen months from target discovery to candidate nomination. These are discovery measurements. Clinical trials, manufacturing work and regulatory review still lie beyond them.
candidate nomination
not total development cost
That boundary is essential. Comparing the price of finding one candidate with the all-in price of bringing a medicine to market would make a wonderful advertisement and a poor piece of accounting. The useful claim is narrower: better prioritization may reduce the time and laboratory work needed to identify something worth developing.
Even the early stopwatch experiment contained rejection. Six synthesized candidates became four biochemical hits, then two cell-validated compounds. The attrition was part of the work. A generated structure is a proposal; synthesis, assays and medicinal chemistry decide whether to keep it.
The billions come with conditions
Insilico's financial model makes this long journey possible. It sells software, licenses therapeutic assets and enters discovery collaborations. In March 2026, its agreement with Lilly included $115 million upfront, an exclusive worldwide license to a portfolio of preclinical oral therapeutics and additional joint research. Conditional milestones could take the deal to approximately $2.75 billion, with royalties beyond that.
The SK Biopharmaceuticals deal in June offered up to $18 million in upfront and near-term milestone payments, against a potential value exceeding $2.5 billion. SK takes responsibility for late-stage development and commercialization of resulting neuroimmune programs. The division of labor is sensible: Insilico supplies discovery expertise; a pharmaceutical partner supplies capabilities farther down the road.
The income statement reveals which engine currently does the heavy lifting. Insilico reported $106.3 million in first-half 2026 revenue, including $103.1 million from drug discovery and pipeline development and $2.70 million from software solutions. It reported $35.54 million in net profit. Large upfront and milestone payments drove the discovery segment, so a profitable half-year should not be mistaken for a smooth subscription curve.
The interruption worth reading
In May 2025, the FDA placed the U.S. Phase IIa rentosertib trial on clinical hold. Insilico's prospectus describes requests concerning safety monitoring and clarifications in study documents, including laboratory findings and related safety considerations. The company updated monitoring, reporting and documents, submitted a response, and reported that the hold was lifted in October.
That episode is part of the company profile, not an inconvenient appendix. The software may accelerate choices made before a trial. Once a compound reaches people, the sponsor still has to meet the requirements of clinical research. Better generation cannot substitute for better monitoring.
What travels beyond the laboratory
There is a practical lesson here for people who will never design a drug. Choose a downstream result that your customer values, then make your tool answer to it. Insilico's published experiments and internal assets expose its platform to measurements beyond a demonstration. Another business can copy that discipline without copying the robot laboratory.
The operating conditions are less portable. This approach needs usable biological data, experimental feedback, medicinal chemistry and enough capital to continue through clinical uncertainty. Sparse or misleading datasets can produce attractive rankings with weak biological foundations. A model can propose a target whose inhibition does not help patients. The eighteen-month discovery figure offers no shortcut through those questions.
Fifty-two weeks is the next unit of progress
On September 10, 2026, Insilico announced the first patient dosed in GENESIS-IPF-3. The Chinese Phase III trial plans 320 participants across 47 centers and 52 weeks of treatment. Its primary endpoint is the annual rate of decline in forced vital capacity, a lung-function measure. Rentosertib remains investigational.
Different studies and objectives. Larger and longer means a stronger test, not a guaranteed result.
September also brought a study applying six proteomic aging clocks to data from the earlier trial. The models indicated reductions in predicted biological age. Those are biomarker findings in patients with lung disease; they do not demonstrate longer lives or an anti-aging treatment for healthy people. Insilico's original longevity ambition has resurfaced, with an intriguing measurement and a considerable distance still to travel.
“From initiating Phase III to obtaining final regulatory approval, it will take three to four years under favorable conditions.”
Zuojun Xu · leading investigator, September 2026
That qualification supplies the right ending. The algorithms have made their suggestions. The laboratory has narrowed them. Now the patients and the calendar will have their say.