AI could help researchers predict which drugs will succeed or fail before clinical trials, potentially saving in development costs and opening the door to more treatments.
Most experimental drugs entering clinical trials never reach patients. Between 2011 and 2020, just 7.9% of drug development programs entering Phase I ultimately received FDA approval. The remaining programs represent years of research, substantial investment, and therapeutic possibilities that never materialized.
Artificial intelligence is already helping researchers identify therapeutic targets and design new medicines, with computationally developed drugs producing successful late-stage clinical results. Now researchers are working toward something potentially more consequential: predicting whether a treatment will work, and what might cause it to fail, before it ever reaches clinical trials.
If AI could identify even a fraction of unsuccessful approaches earlier, pharmaceutical companies could preserve development capital and pursue more therapeutic opportunities. The potential economic impact extends well past faster drug discovery.
AI Drug Discovery Is Already Delivering Results
Computational methods are already helping researchers analyze protein structures, predict molecular interactions, and design compounds against therapeutic targets.
Takeda’s zasocitinib, an oral TYK2 inhibitor, demonstrated efficacy in Phase III psoriasis trials, including a head-to-head study against the approved treatment Sotyktu. In September 2026, the FDA accepted its New Drug Application under Priority Review. Its discovery relied heavily on physics-based, structure-guided molecular modeling, an approach distinct from newer generative AI methods.
Insilico Medicine has incorporated AI into both target identification and molecular design. Its PandaOmics platform analyzed biological data to identify TNIK as a potential therapeutic target for idiopathic pulmonary fibrosis, a progressive disease characterized by lung scarring. Researchers then used Insilico’s generative chemistry platform to develop rentosertib, an oral TNIK inhibitor.
In a randomized Phase IIa trial involving 71 patients, rentosertib produced favorable lung-function results compared with placebo after 12 weeks. Insilico dosed its first Phase III patient in September 2026.
Researchers are now investigating whether AI can help anticipate what happens after a drug acts on its target, including changes that could determine whether a treatment succeeds or fails.
Could AI Predict Clinical Failure Before It Happens?
Researchers are developing AI models, often called virtual cells, designed to predict how living cells respond to experimental treatments. Take cancer, for example. Tumors can develop resistance to a drug by activating alternative biological pathways that allow them to survive. A predictive model could anticipate those responses and help researchers identify combinations of drugs capable of overcoming resistance before testing them in the laboratory.
An October 2026 study in Nature Biotechnology examined AI models designed to predict how cells respond when specific genes are altered. Researchers found that several models captured biological changes that conventional measures of accuracy had overlooked. The findings suggest that AI can predict certain cellular responses, an important step toward anticipating how experimental treatments might affect disease.
Improving these predictions requires more data on how cells respond to experimental interventions. In October, Biohub and its partners announced a $1.8 billion expansion of the Virtual Biology Initiative to generate large-scale datasets documenting cellular responses to genetic changes, drugs, and other experimental conditions. The data could help train AI to predict how cells respond to new treatments.
What Would Better Predictions Mean for Drug Development?
In BIO’s analysis of drug development programs, only 28.9% of candidates entering Phase II advanced to Phase III. An August 2026 analysis in Nature Reviews Drug Discovery found that improving Phase II success rates could substantially reduce the cost of bringing a successful medicine to market.
Predictive biology could help researchers recognize potential failures earlier by anticipating how treatments affect disease pathways, identifying resistance mechanisms, and revealing unexpected responses. These insights could strengthen the evidence supporting a drug candidate before companies commit to costly clinical development.
Identifying unsuccessful drugs earlier could change how pharmaceutical companies spend their research budgets. Capital that might otherwise be spent on unsuccessful programs could support additional drug candidates, expand research pipelines, or make treatments for smaller patient populations more financially viable.
Even narrowly focused predictive models could become valuable research tools. A system that reliably anticipates a specific biological response could reduce the experiments and investment needed to investigate a therapeutic hypothesis, freeing researchers to explore more possibilities with the same resources.
The Industry Is Investing in Predictive Biology
In October, AstraZeneca announced a collaboration with insitro involving $40 million upfront and approximately $1.46 billion in potential milestone payments. The partnership will combine AI, human biological data, and laboratory experiments to investigate disease mechanisms and develop treatments for liver and heart diseases.
As AI models become more accessible, specialized biological data and experimental capabilities could become increasingly valuable. Companies that generate their own data can test predictions and refine their models, while initiatives like Biohub’s open datasets could make the underlying technology available to a wider range of researchers.
For investors and pharmaceutical partners, the value of these platforms will depend on how accurately they predict experimental results and whether those predictions save researchers time and money. Companies that consistently deliver those results could gain an advantage as predictive biology becomes more widely adopted.
The Value of Knowing Earlier
Clinical trials will remain essential for establishing whether a treatment is safe and effective in patients. The opportunity for predictive biology lies in what researchers could learn before reaching that stage.
If predictive models can reliably challenge weak therapeutic hypotheses, uncover unexpected biological responses, and guide researchers toward more productive experiments, companies could enter clinical development with stronger evidence supporting their candidates.
In an industry where failure is both common and extraordinarily expensive, learning earlier which ideas deserve to advance could be as consequential as discovering new ones.