Virtual Patients Are Predicting Real Drug Failures—And Pharma Is Taking Notice

The pharmaceutical industry's $140 billion annual investment in clinical trials yields an uncomfortable truth: 88% of drugs tested in humans ultimately fail. Now, artificial intelligence companies claim they can predict these expensive failures before patients ever enroll—and early results suggest they may be onto something.

BioinvestGPT, a Copenhagen-based startup founded in 2024, has correctly predicted the outcomes of five out of six high-profile clinical trials by creating virtual patients and running simulated studies. The company's track record includes forecasting Novartis' del-desiran failure for muscular dystrophy—a result that wiped $30 billion from the company's market value—and Moderna-Merck's melanoma vaccine success, according to The Independent.

How Virtual Trials Work

The technology behind these predictions relies on DNA sequencing to model virtual human bodies that match specific trial eligibility criteria. BioinvestGPT then runs simulated trials with digital drug models, completing analyses in one month or less—a stark contrast to the years required for traditional Phase 1-3 trials.

The company's founders, Bragi Lovetrue and Idonae Lovetrue, have demonstrated their system's capabilities across multiple therapeutic areas. Beyond the Novartis and Moderna-Merck predictions, BioinvestGPT accurately forecast weak clinical benefit for AstraZeneca-Ionis' heart drug Wainua, Novo Nordisk's initial ziltivekimab failure, and Vaxcyte's pneumococcal vaccine success.

However, the technology isn't infallible. BioinvestGPT predicted success for Novartis' pelacarsen, a lipoprotein(a) reducer, which ultimately failed in September trials—highlighting current limitations in modeling complex genetic variations.

Industry Adoption and Skepticism

Investment in AI drug discovery reached $8.4 billion in 2025, more than doubling from 2023 levels, according to a McKinsey report cited by The Independent. Yet pharmaceutical executives remain cautious about relying on predictive algorithms for go/no-go decisions.

Tel Aviv-based competitor QuantHealth takes a different approach, combining AI with real-world data. The company's Chief Medical Officer, Francisco Beca, suggests that by 2027-2029, it may become unethical to proceed with trials that AI predicts will fail—raising profound questions about patient enrollment in studies with high predicted failure probability.

McKinsey partner Alex Devereson notes that pharma companies are currently "dipping their toes" into AI trial modeling, with most AI spending still concentrated on molecule design rather than efficacy demonstration.

What's Next

BioinvestGPT has issued new predictions that will test its credibility: the company forecasts failure for Biogen's litifilimab in Phase 3 lupus trials and Takeda's zasocitinib in Phase 2 studies for Crohn's disease and ulcerative colitis. Industry leaders from both companies, including Takeda's Andy Plump and Biogen's Diana Gallagher, have expressed caution about definitive predictions.

The timing may be opportune for AI integration. U.S. health regulators announced initiatives last week to accelerate drug trials, potentially creating pathways for incorporating predictive AI into the approval process. As the technology matures, it could fundamentally alter how pharmaceutical companies allocate capital and assess M&A targets—transforming an industry where only 12% of drugs tested in humans ultimately gain regulatory approval.

Sources

1. AI companies see success in predicting drug failures after virtual human trials — The Independent