Key points
- Africa needs to own and develop AI tools trained on genomic data generated from its populations, an infectious disease expert says.
- Models trained mainly on data from outside the continent may perform poorly when applied to African populations and pathogens.
- Experts also called for better genomic metadata and more sustainable public health financing.
Main story
Africa needs to strengthen its capacity to analyse genomic data and develop artificial intelligence tools locally if it is to improve disease surveillance and public health outcomes, an infectious disease expert, Dr Oladipo Kolawole, has said.
Kolawole, an Associate Professor of Infectious Diseases and Founder of Helix Biogen Institute, Ogbomoso, made the call on Monday at the fourth Annual Scientific Symposium organised by the Institute of Human Virology Nigeria (IHVN) in Abuja.
The symposium, themed “Shaping the Future of Public Health Landscape through Sustainable Funding, Artificial Intelligence, Machine Learning, and Innovation,” examined the role of emerging technologies in strengthening public health systems.
Kolawole said Africa would struggle to develop effective genomic intelligence if artificial intelligence models continued to depend largely on datasets generated from populations outside the continent.
He said combining Whole Genome Sequencing (WGS) with machine learning could help African countries anticipate disease trends and antimicrobial resistance, but the accuracy of such systems would depend heavily on the data used to train them.
According to him, differences in genetic and pathogen profiles mean that AI models developed using data from other populations may not perform equally well when applied to African settings.
He cited antimicrobial resistance prediction as an example, saying a global AI model that predicted ciprofloxacin resistance with about 75 per cent accuracy in England achieved only about 50 per cent accuracy when applied to African data.
Kolawole said the gap demonstrated the need for researchers on the continent to develop models using locally generated genomic information.
“Models trained elsewhere do not automatically generalise to African populations due to distinct lineages and underrepresentation of our pathogen diversity.
“One size does not fit all; we must own our data and our AI tools,” he said.
The expert, however, said African researchers should not wait until comprehensive genomic datasets were available before developing local AI capacity.
He advised researchers to begin with available high-quality data and use it to establish foundational models that could be strengthened as additional genomic information and metadata became available.
Kolawole also addressed concerns about the potential misuse of advanced biotechnology and AI, saying biosecurity safeguards were increasingly being incorporated into research environments.
He said researchers and students at his institute were required to sign biosecurity agreements before beginning their engagements, stressing that those developing and using such technologies must remain legally and ethically accountable.
The session chairperson, Prof. Alash’le Abimiku, reinforced the importance of the quality of data used in genomic research.
Abimiku recalled challenges during COVID-19 research, when missing metadata made it difficult to connect genomic information with patients’ clinical profiles.
She urged researchers to make comprehensive metadata collection and quality control an integral part of genomic data gathering from the beginning.
Representatives of eHealth Africa, KNCV Tuberculosis Foundation and APIN Public Health Initiatives commended IHVN for sustaining the scientific symposium.
They also called for a shift away from temporary, emergency-driven approaches to public health financing towards sustainable institutional infrastructure that could allow health systems to respond more quickly to future crises.
The issues
The central issue is not simply whether Africa can use AI in public health, but whether it can control the data and models on which those AI systems depend.
The accuracy gap cited by Kolawole shows the potential consequence of relying on models trained predominantly with data from other populations. If African genetic and pathogen diversity is poorly represented, tools developed elsewhere may not perform reliably when applied to African health challenges.
The quality of the underlying data is another challenge. Abimiku’s experience during COVID-19 highlights how missing metadata can limit the ability of researchers to connect genomic findings with clinical information.
This makes local data generation, accurate metadata collection and research capacity important components of Africa’s ability to build useful genomic intelligence.
What’s being said
Kolawole said:
“One size does not fit all; we must own our data and our AI tools.”
He also said models trained outside Africa did not automatically generalise to African populations because of differences in pathogen lineages and the underrepresentation of the continent’s pathogen diversity.
Abimiku stressed the need for researchers to incorporate comprehensive metadata documentation and quality control into genomic data collection from the outset.
What’s next
African researchers are expected to continue developing genomic datasets and foundational AI models using available high-quality local data, while improving those systems as more genomic information and metadata become available.
Research institutions will also need stronger data-quality and biosecurity practices, while public health stakeholders are calling for more sustainable financing and institutional infrastructure.
Bottom line
Africa’s ability to use AI effectively in disease surveillance will depend partly on whether it can generate and control the data used to train its models.
For Kolawole, the path to more reliable genomic intelligence is clear: African researchers need to build from African data rather than assume that models developed elsewhere will adequately capture the continent’s biological diversity.
