A Copenhagen-based artificial-intelligence startup called BioinvestGPT correctly forecast five of six high-profile drug-trial outcomes that it shared with Reuters in July, before the results were public.
That is a more important claim than a backtest. Reuters says it received the company’s analyses before the readouts and could therefore compare the predictions with the eventual clinical results. Five matched. One did not.
Five Correct Calls, One Miss
Reuters reports that BioinvestGPT correctly called a negative result for Novartis’ del-desiran, success for Moderna and Merck’s personalized melanoma vaccine, weak clinical benefit for AstraZeneca and Ionis’ Wainua, the first failed trial for Novo Nordisk’s ziltivekimab, and success for Vaxcyte’s pneumococcal vaccine.
The miss was Novartis’ pelacarsen. BioinvestGPT expected a positive result, but the Phase 3 study failed to reduce the risk of major cardiovascular events in the target population. Co-founder Bragi Lovetrue told Reuters that the model had failed to account correctly for genetically determined variation in lipoprotein(a) particle size.
The System Runs Virtual Clinical Trials
BioinvestGPT says its BVCT system uses genomic information and mathematical models to simulate a human body matching a trial’s eligibility criteria, then estimates how that virtual population will respond to a drug. That puts the system closer to a computational virtual biotech than a conventional chatbot making predictions from headlines.
The company maintains a public prediction dashboard covering hundreds of trials. Its own frozen September 24 cohort claims 674 correct calls out of 705 scored readouts, or 95.6%. That broader record is company-reported; BioinvestGPT says the prediction documents are timestamped, but the underlying sealed documents are not all public. The Reuters five-of-six subset is more independently useful because the newsroom received the calls before the outcomes were announced.
Del-Desiran Was the Eye-Catching Prediction
Novartis had positioned del-desiran as a potentially important treatment for myotonic dystrophy type 1. Reuters says the company’s CEO had previously discussed peak annual sales of roughly $5 billion. BioinvestGPT predicted in July that the late-stage trial would show insignificant clinical benefit.
The trial later missed its primary endpoint, and Novartis shares fell 11%, erasing roughly $30 billion in market value. BioinvestGPT’s own trial dashboard records the del-desiran prediction as a true negative made 44 days before the September readout.
Why Drug Companies Care
Human clinical trials are expensive and slow. Reuters reports that the global biopharmaceutical industry spends roughly $140 billion a year on human clinical testing while only about 12% of drug candidates ultimately win regulatory approval.
If virtual trials can reliably identify weak candidates before large Phase 2 or Phase 3 studies begin, companies could avoid some expensive failures and focus capital on treatments with stronger predicted clinical benefit. That does not make physical trials optional. It changes which drugs might deserve to reach them.
This Is Also a Test of AI in Biology
BitcoinVersus has already covered AI systems moving deeper into biology, including Claude identifying an enzyme system with CRISPR-like DNA repeats. BioinvestGPT pushes the idea further downstream: instead of helping discover a biological mechanism, it is trying to forecast whether a therapy will survive the clinical-validation stage.
That is a harder benchmark because real trial outcomes reflect biology, patient selection, dosing, endpoints, statistics and trial design at the same time. Going five-for-six does not prove the system can predict medicine. It does show that prospective AI-based trial simulation has moved beyond purely hypothetical demonstrations.
The Next Public Test Is Already Lined Up
Reuters says BioinvestGPT is now predicting failure for two Phase 3 studies of Biogen’s litifilimab in systemic lupus erythematosus and for Phase 2 studies of Takeda’s zasocitinib in Crohn’s disease and ulcerative colitis. Those results are expected before the end of 2026.
That gives the technology something AI claims often lack: a dated, falsifiable next test. If those trials report, the predictions can be checked against the real outcomes rather than judged only through demos or retrospective benchmarks.
BitcoinVersus.Tech
Advertisement
Editor’s Note
The five-of-six result refers specifically to the high-profile predictions BioinvestGPT shared prospectively with Reuters in July 2026. BioinvestGPT’s larger dashboard accuracy figures are company-reported and should not be treated as an independently audited clinical-performance measure.
Support and donation options are available through BitcoinVersus.Tech.
BitcoinVersus.tech is not a financial advisor. Content is provided for informational purposes.
Leave a comment