Future Explained

An AI Shortcut Just Cut Drug-Testing Time by 94%. Could It Make Drug Manufacturing Cheaper Everywhere?

Brown University's new AI model cuts drug-formulation testing time by up to 94%, precisely the stage where manufacturers in Africa, Asia and beyond already operate but cannot yet compete.

Illustrative photograph, not Brown's own lab. Photo: Brown University.

A drug that releases slowly over hours or days, a nicotine patch, an extended-release tablet, a wound-healing bandage laced with medicine, has to be tested extensively before regulators will approve it, because the manufacturer must prove exactly how fast the drug leaves the material once it is in the body. That testing is normally slow and expensive, run in real time over the full length the product is meant to work. On 6 July 2026, engineers at Brown University led by Vikas Srivastava published a physics-informed AI model in the Journal of Drug Delivery Science and Technology that predicts the same long-term release behaviour from a small fraction of that data, a shortcut that matters most to manufacturers who cannot afford the years of testing it currently takes.

How does an AI model shortcut years of drug testing?

The model, built by Srivastava with Daanish Qureshi and Khemraj Shukla, combines short bursts of real experimental measurements with Fick's Law of Diffusion, the century-old physics equation describing how a substance spreads through a material, inside a structure known as a physics-informed neural network, a method originally developed by Brown mathematician George Karniadakis. For simple, flat materials, the model needed only 6% of the experimental data a conventional study would require to predict the same long-term release curve, cutting testing time by 94%. For more complex materials, ones with folds or wrinkles that change how a drug diffuses through them, it still needed only a third of the data, a 67% reduction. "In pharmaceutical development, time is money," Srivastava said. "We're hopeful that this approach can help in getting products to patients more quickly and less expensively." The same underlying approach, the researchers say, applies in principle to pills and other controlled-release systems as well as topical patches.

Why would this matter more in Lagos or Nairobi than in Providence?

Because the testing bottleneck Brown's model shrinks sits exactly where African pharmaceutical manufacturing already operates and struggles. Africa imports more than 70% of the medicines it uses and just 1% of the vaccines administered on the continent are produced locally, according to Africa CDC, which has set a target of 60% local production of vaccines, diagnostics and therapeutics by 2040 through initiatives including the Platform for Harmonised African Health Products Manufacturing and the African Medicines Agency. Around 600 pharmaceutical manufacturers operate across the continent, but 80% are concentrated in just eight countries, and most of them work only at the formulation and packaging stage, buying in active ingredients rather than making them, exactly the layer of drug development where Brown's model applies. Those plants also run at 30% to 60% of capacity, well below the 70%-plus typical of manufacturers in wealthier countries, with production costs that remain higher than equivalent plants in China or India. A tool that cuts the time and cost of formulation testing addresses a real constraint at the one stage of the pharmaceutical supply chain African manufacturers are already positioned to compete in.

Is anyone actually trying to close this gap today?

Yes, though with a different technology solving an adjacent problem. The University of Cape Town's H3D Centre in South Africa is piloting a reactor that manufactures active pharmaceutical ingredients, the raw drug compounds themselves, in multi-kilogram batches far faster than conventional chemistry, in a one-year, $700,000 project funded through USAID's MATRIX initiative and developed with the US-based Oak Crest Institute of Science. The pilot targets antiretrovirals and HIV prevention drugs, chosen partly because South Africa imports 100% of its HIV-related active ingredients despite having its own formulation capacity, a gap the project is designed to close. It uses a different technology from Brown's AI model, tackling ingredient production rather than formulation testing, but it shows the same underlying logic already being tested on the continent: that shortening an expensive, slow step in drug manufacturing can shift where medicines actually get made.

What would it take for a tool like this to actually reach African or Asian manufacturers?

Several things that have not happened yet. Brown's model has been validated on the materials tested in Srivastava's own lab; it would need retesting on the specific excipients, climate conditions and manufacturing processes used by manufacturers elsewhere before anyone could rely on its predictions. It would need acceptance from regulators, including the World Health Organization's prequalification programme, which sets the bar most African and Asian generic manufacturers must clear to sell into public health markets. And it would need someone, a university partnership, a donor programme, or the researchers themselves, to actually build the bridge between a Rhode Island engineering lab and a formulation plant in Kenya, Ghana or India. None of that exists today. Whether the 94% figure becomes a tool a Nairobi or Ahmedabad manufacturer actually uses, or stays a result published in an American engineering journal, is the question this research raises without yet answering.

Institutions in this article: Brown University; University of Cape Town, H3D Centre; Africa CDC.

Frequently Asked Questions