AI FOR SCIENCE · INTERVIEWS
Aliper says Insilico's platform brings 'zero to preclinical candidate' down to 12 to 18 months on average at orders of magnitude lower cost; it has nominated 33 candidates, 13 programmes have reached or been cleared for human trials, and the lead lung-fibrosis drug rentosertib is in Phase 3, which he calls the first AI-discovered drug to get there.
The record has rentosertib's Phase 2a and first Phase 3 dosing, but no pipeline-wide counts, no average timeline, no pain drug and no new AI strategy. Here he gives the 33 and 13, the 12-to-18-month average, a non-opioid pain drug that is preclinical only, and the name of a strategy, 'pharmaceutical superintelligence frontier'.
All of it is Aliper on his own company, in 13 minutes: no project list sits behind the 12 to 18 months, the cost has no figure, the pain drug's 'better by 50% and better than morphine' is preclinical with no animal, dose or comparator given, and the new AI strategy has no detail.
Aliper says going from zero to a preclinical candidate traditionally takes four and a half to five years; Insilico built a platform for every stage of discovery, and its average is now 12 to 18 months at orders of magnitude below hundreds of millions of dollars. He says studies have shown the timeline 'over and over'. Once it is cheap, indications that did not pay before can be pursued.
He says the most advanced programme is idiopathic pulmonary fibrosis (IPF): just in Phase 3 after Phase 2a, where the top dose, 60 mg once a day for 12 weeks, did not lose lung capacity as the placebo group did but regained about 100 mL of forced vital capacity (FVC). On that basis he calls it, 'to my knowledge', the first AI-discovered drug to reach Phase 3 and the most advanced AI-discovered therapeutic in the world.
He counts by stage: fibrosis in Phase 3, immunology in Phase 2, oncology in Phase 1, plus cardiovascular and metabolic work and neuroscience. The IBD drug is an oral molecule that repairs the gut barrier and is not absorbed into the blood. The pain drug was announced recently, with a novel target and mechanism and no addiction liability; he says it showed 50% better efficacy than the most advanced current pain drug (the captions read 'NAF 1.8', probably Nav1.8) in preclinical work, its injectable form beat morphine in preclinical models, and it is in IND-enabling studies, with clinical trials hoped for 'sometime soon'.
He says the company has just announced a new AI strategy, 'pharmaceutical superintelligence frontier', orchestrating every task through foundation models, large language models and specialist models. At the same time clinical trials cannot cut corners and the regulatory process stands; claims of curing all diseases in five years are unrealistic unless the regulatory path changes. His yardstick is that the FDA approves about 50 drugs a year, so doubling would already be big, then 100, 200 and perhaps, one day, 1,000.
“Claims that say that we can cure all diseases in like 5 years, unless we change the regulatory path, it's just not realistic.”Alex AliperAsked about that day's panel, he says one data type can be very valuable for a task and another not, and many diseases need therapies designed for a specific population. His example is the 30th preclinical candidate nominated this year, the first discovered in Abu Dhabi: an oncology drug against tumours carrying a gene deletion (the captions read 'mab deletion', probably MTAP), working by synthetic lethality so that it kills those tumour cells in a targeted way.
The timeline records rentosertib's Phase 2a as randomised, double-blind, 71 patients, 12 weeks, with safety the primary endpoint and lung function a secondary one; the first Phase 3 dosing on 10 September, planned for 320 people at 47 centres in China over 52 weeks.
Aliper tells the Phase 2a as the top dose group 'regaining about 100 mL of FVC' and calls the drug the first AI-discovered one in Phase 3; he does not mention that safety was the primary endpoint, or where the Phase 3 runs and how large it is.
The timeline and /funding carry Insilico's Hong Kong listing and several rentosertib entries; they hold no pipeline-wide count, no average time to a candidate, and no pain or IBD drug.
33 nominated candidates, 13 programmes in or cleared for human trials, an average of 12 to 18 months; a non-opioid pain drug that is preclinical only; and a 30th candidate, from Abu Dhabi, given only in words.
The record holds no new Insilico AI strategy.
He says the company has just announced a 'pharmaceutical superintelligence frontier' strategy that orchestrates every task through foundation, language and specialist models; he gives no model, timetable or measure.
It measures zero to a preclinical candidate, not to a drug, and Aliper himself says regulation and trials cannot be shortened. The number is therefore better set against the stretch after it than against the total time to a medicine. That is our reading.
Phase 3 has only begun and nothing in it is measured; the 100 mL he cites comes from a 12-week Phase 2a whose primary endpoint was safety. Read the two apart, and 'reached Phase 3' does not become 'works'.
Zero to a preclinical candidate takes 12 to 18 months on average, at orders of magnitude lower cost than the traditional route.
Company-reported
No project list, dates or cost figure; he says studies have shown it 'over and over' without naming one.
33 developmental candidates are nominated and 13 programmes have reached or been cleared for human trials.
Company-reported
He puts it two ways, 'progressed into human trials' and 'received IND approval and clearance', which are not quite the same thing.
Rentosertib has just entered Phase 3 after Phase 2a, where the top dose regained about 100 mL of FVC after 12 weeks while placebo declined.
Company-reported
Insilico's own release for the first Phase 3 dosing is on the timeline; the 100 mL is his rounded figure, and this page did not reread the Phase 2a paper.
It is the first AI-discovered drug in Phase 3 and the most advanced AI-discovered therapeutic in the world.
Guest's judgment
He adds 'to my knowledge'; it is a ranking claim and this page does not test it.
A gut-restricted oral IBD drug repairs the gut barrier and is not absorbed into the blood.
Company-reported
A description of mechanism with no data; he does not give its stage.
A non-opioid, non-addictive pain drug with a novel target and mechanism shows 50% better efficacy than the most advanced pain drug in preclinical work, and its injectable form beats morphine.
Company-reported
No data, species, dose or comparator beyond the thing the captions render as 'NAF 1.8', probably Nav1.8.
This year it nominated its 30th preclinical candidate, the first discovered in Abu Dhabi, against tumours with a gene deletion and acting by synthetic lethality.
Company-reported
The target is not named (the captions suggest MTAP) and there is no data.
A new AI strategy, 'pharmaceutical superintelligence frontier', orchestrates every task through foundation, language and specialist models.
Target
No model, timetable or measure.
Curing all diseases in five years is not realistic unless the regulatory path changes; the FDA approves about 50 drugs a year and doubling would already be big.
Guest's judgment
His view and estimate; the 50 is his approximation.
Whether the 12-to-18-month figure is tied to named programmes and dates.
Would show up as A list from Insilico of its candidates with the start and nomination date of each, ideally with how cost is counted.The primary-endpoint readout of the Phase 3, GENESIS-IPF-3.
Would show up as The 52-week FVC decline from the sponsor or the trial registry; the Phase 2a's 100 mL cannot stand in for it.The pain drug's preclinical data and its IND filing.
Would show up as A preprint or poster naming the target, species, dose and comparator, or a regulator's clearance to start a trial.What the 'pharmaceutical superintelligence frontier' strategy becomes.
Would show up as Insilico's own announcement naming the models, the targets it will meet and the dates.Read 2026.10.14 The full timestamped captions pasted by the author (machine-made, no speaker labels; only the host, who is not named, and Aliper speak), read in full; the video and audio were not watched. It is a short on-site interview on the conference organiser's own channel. The captions garble two terms: 'NAF 1.8' is read here as the pain target Nav1.8 and 'mab deletion' as an MTAP deletion; both are guesses and were not checked. Open the interview ↗