Key facts
- AI simulations are being used to identify promising pharmaceutical programs and assess acquisition targets.
- The global biopharmaceutical industry spends $140 billion annually on human clinical testing, with only a 12% approval rate for drug candidates.
- BioinvestGPT has correctly predicted the outcomes of five out of six high-profile drug trials.
- QuantHealth uses AI and real-world data to simulate patient responses to therapies.
- Biogen's litifilimab and Takeda's zasocitinib are predicted to fail in upcoming trials, according to BioinvestGPT.
- Human trials will remain necessary, but AI simulations can help assess experimental drug potential.
AI companies are developing virtual drug trials to predict the success of human studies, aiming to reduce the high failure rate in pharmaceutical research. Novartis' experimental drug del-desiran for a rare type of muscular dystrophy, which had been predicted to achieve $5 billion in peak annual sales, missed its late-stage trial goal last month, causing its shares to fall 11% and erasing $30 billion in market value.
BioinvestGPT, an AI startup, was not surprised by the outcome, having run a simulated trial in July that predicted insignificant clinical benefit for del-desiran. The company and others like it are working with drugmakers on virtual trials to increase the likelihood that experimental medicines will prove safe and effective in actual patients, thereby avoiding costly traditional clinical trials.
The global biopharmaceutical industry spends approximately $140 billion annually on human clinical testing, yet only about 12% of drug candidates receive regulatory approval, a rate that has remained stagnant for decades. AI firms assert that their simulations can help identify which pharmaceutical programs are worth pursuing and evaluate potential acquisition targets. As AI tools become more adept at flagging clinical risks, unexpected trial failures are expected to decrease.
Human clinical trials typically progress through Phase 1 (safety), Phase 2 (mid-stage), and Phase 3 (large-scale efficacy) studies, a process that can take years. In contrast, some AI simulations can be completed in a month or less. Francisco Beca, chief medical officer at QuantHealth, emphasized the need to reduce trial failures rather than solely focusing on speed.
Investment in AI drug discovery more than doubled to $8.4 billion in 2025 compared to 2023, according to McKinsey. The firm noted that current spending is concentrated on areas where AI excels, such as molecule design, rather than on major industry bottlenecks like proving drug efficacy through lengthy trials. McKinsey partner Alex Devereson stated that pharmaceutical companies are cautiously exploring AI trial modeling to assess drug candidates before committing capital.
US health regulators recently announced initiatives to accelerate drug trials, which could potentially create a pathway for predictive AI in clinical development. BioinvestGPT has shared detailed analyses of several high-profile drug trials with Reuters, correctly predicting outcomes in five out of six cases, including the negative result for del-desiran, the success of Moderna and Merck's melanoma vaccine, and the weak benefit of AstraZeneca and Ionis' heart drug Wainua.
Bragi Lovetrue, co-founder of BioinvestGPT, explained that the platform uses DNA sequencing to simulate human bodies matching trial eligibility criteria and then models the test drug's performance. While the simulations are not always accurate, as seen with Novartis' pelacarsen, where the mechanism was misunderstood, they offer valuable insights. QuantHealth has published simulations for ulcerative colitis and cholesterol drug trials.
BioinvestGPT also predicts failure for two Phase 3 trials of Biogen's litifilimab for lupus and for Takeda's zasocitinib in Crohn's disease and ulcerative colitis, citing suboptimal drug suitability for the specific patient populations. Takeda's research chief, Andy Plump, expressed confidence in the drug's mechanism but doubted AI's current ability to make definitive predictions. Biogen's head of clinical development, Diana Gallagher, acknowledged the use of AI but noted that algorithms relying on historical data might predict negative results for lupus drugs, for which few treatments exist.