UCT Wins $1.6 Million For African Drug-Response AI

The University of Cape Town has secured US$1.6 million from Google.org to develop an artificial-intelligence platform focused on a long-standing weakness in medical evidence: the limited representation of African populations in the data used to understand how people respond to medicines.

The AfriPGx-AI project will map genetic variation across more than 50 genes involved in drug metabolism. Computational models will predict which variants are most likely to change how a medicine is processed, and the strongest predictions will then be tested in the laboratory. The first phase will concentrate on treatments for tuberculosis, HIV and malaria.

According to UCT's announcement, the grant is worth approximately R26.4 million. UCT was one of 17 organisations selected from more than 3,000 applicants across 30 countries for the Google.org Impact Challenge: AI for Science. The selected organisations will share US$36 million.

The project is led through UCT's H3D drug-discovery centre and brings together pharmacology, computational biology, mathematical modelling and artificial intelligence. Professor Kelly Chibale, Dr Musalula Sinkala, Dr Mwila Mulubwa, Dr Mathew Njoroge and Professor Jonathan Shock are among the researchers named in the programme.

The scientific problem is significant. Many pharmacogenomic guidelines have been developed primarily from European and East Asian data. Africa contains the greatest human genetic diversity, yet clinicians on the continent often make prescribing decisions without evidence that captures that variation. A dose considered standard elsewhere may be processed differently by particular populations, affecting efficacy or the risk of adverse reactions.

AI can help narrow a very large search space by identifying variants that deserve laboratory attention. But the laboratory validation is what gives the project credibility. A prediction alone cannot establish that a variant changes drug metabolism. AfriPGx-AI's proposed feedback loop between computation and experiment is designed to move beyond pattern recognition toward evidence that can inform treatment.

The project also promises to release its software, AI models and curated datasets openly. That commitment matters because African scientific institutions are often asked to supply data to projects whose resulting tools remain controlled elsewhere. Open outputs can allow researchers to reproduce findings, extend the methods and apply them to other diseases or populations.

Governance will still require care. Genetic information is sensitive and can expose families and communities, not just individual participants. Consent, secure data handling and rules for secondary use must be treated as part of the research design. Open science does not mean releasing identifiable health data.

UCT says the resulting decision-support tools will be designed for low-bandwidth environments. That is a practical requirement if the work is to reach clinics beyond major research hospitals. It also means interfaces and recommendations must be understandable to health professionals who may not have specialist training in genomics.

The initiative complements UCT's wider role in African AI research, including its hub on AI safety, peace and security. In both cases, the value lies in building local scientific capacity rather than treating Africa only as a deployment market.

AfriPGx-AI will need time before it can influence clinical guidelines. Its immediate contribution is more foundational: creating African-led, openly shared evidence about populations that have been underrepresented in medical research. If its computational predictions survive laboratory testing, the project could help move precision medicine closer to the patients who have historically benefited least from it.

Source: University of Cape Town.