New paper shows how AI can accelerate drug discovery in low-resource computational settings

26 September 2023 | Story Helen Swingler. Photo Je’nine May. Read time 6 min.
UCT’s H3D centre, where a virtual screening cascade was developed for malaria and TB drug discovery.
UCT’s H3D centre, where a virtual screening cascade was developed for malaria and TB drug discovery.

Artificial intelligence (AI) and machine learning (ML) promise to catapult African drug discovery into a new era, a new research paper in Nature Communications has demonstrated. Co-authored by University of Cape Town (UCT) scientists, the paper describes how an AI/ML-based modelling tool, which requires low computational resources, mined an existing drug discovery database to predict new medicinally active compounds – before these were synthesised in the laboratory.

The paper describes the first-of-its-kind fully automated AI/ML virtual screening cascade implemented at a drug discovery centre in Africa. Using in-house data collected over a decade by UCT’s Holistic Drug Discovery and Development (H3D) Centre, a virtual screening cascade was developed for malaria and tuberculosis (TB) drug discovery.

This produced 15 models for key decision-making screening assays (investigation in the laboratory). The scientific term used to describe the computer or virtual simulations that investigate pharmacological hypotheses is “in silico”, which refers to experimentation performed by computer.

The research paper is the first from a partnership between the H3D and the Ersilia Open-Source Initiative (EOSI). EOSI is a United Kingdom-based non-profit organisation specialising in AI and ML that exploits existing drug discovery datasets in the quest for new drugs that address urgent biomedical needs in low- and middle-income countries (LMICs).


“Through this work we have brought AI directly to production, addressing a key concern.”

The authors describe the deployment of the EOSI-developed modelling tool, ZairaChem. Fully automated, the tool works across a broad spectrum of datasets and has been designed for fast and easy implementation in settings with scarce computational resources. Strong data science skills are not required.

The joint first authors are Dr Gemma Turon (chief executive officer and co-founder, EOSI) and Dr Jason Hlozek (postdoctoral fellow at H3D). Dr John Woodland (research officer at H3D) and Dr Ankur Kumar (EOSI) are the other co-authors. The joint senior lead authors are Dr Miquel Duran-Frigola (lead scientist and co-founder, EOSI) and Professor Kelly Chibale, the Neville Isdell Chair in African-centric Drug Discovery and Development who is also the founder and director of H3D.

Malaria and TB drugs targeted

Professor Chibale is upbeat about prospects.

“This work serves as proof-of-concept for the potential of AI and ML tools to support drug discovery in low- to medium-income countries, including those in Africa,” he said.  

While the paper shows how compounds can be computationally profiled prior to synthesis and testing and can inform the progression of frontrunner compounds at H3D, it is also the first instance of a virtual screening cascade built with data produced on and for Africa.

Key benefits of this technology/development for the continent include:

  • generating AI capacity based on in-country data
  • aiding all the stages of drug discovery and development and reducing the risk of failures
  • addressing the significant problem of under-exploitation of data produced in African drug discovery programmes because of the limited capacity to develop and deploy AI models
  • significantly reducing the costs of drug discovery in Africa, which can be a game-changer in the resource-constrained setting of most African countries.

“Through this work we have brought AI directly to production, addressing a key concern in a field where so much AI research is being done but only a small percentage is eventually implemented in real-world scenarios,” Dr Duran-Frigola noted.

This work is timely in view of the recent initiative by global charitable foundation, the Wellcome Trust. Chibale was a member of a scientific advisory committee of some 15 global leaders in drug discovery across industries and geographies convened by the trust to contribute and guide research that would inform the recently published landscaping report, “Unlocking the potential of AI in drug Discovery”, led by Boston Consulting Group.

The H3D–EOSI collaboration is highlighted in this report, within the context of addressing barriers such as capability gaps that must be addressed to unlock the full potential of AI, particularly in LMIC settings.

This report set out to understand where AI is currently being deployed, identify promising new areas to expand its use, and assess the adoption of AI tools across the drug discovery ecosystem. Another strong focus was to understand where AI can be expanded across infectious disease therapy areas and how to increase adoption in LMICs.


“Computer and pharmaceutical sciences have for a long time been siloed. AI can bring them together.”

There are other positive developments allied to this development.

“At a broader level, AI applied to drug discovery paves the way for introducing computer scientists – a sector that is blooming in Africa – into health sciences,” said Chibale.

Duran-Frigola added, “Computer and pharmaceutical sciences have for a long time been siloed. AI can bring them together and, most importantly, find a purposeful application of the growing community of programmers and data scientists.”

High costs minimised

The prospects are also enticing when evaluated against the cost of bringing new medicines from the bench to the bedside. According to the paper’s authors, these costs have risen steadily since the 1970s. Recent estimates suggest a median cost of US$1.3 billion per drug with research and development taking an average of 10 years.

The drug discovery industry is now harnessing AI and learning to avoid costly failures in drug development and accelerate research timelines and reduce attrition rates.

“Investments in AI and ML have also soared in the past five years,” said Chibale.

There is also the promise of extending AI and ML to the field of infectious diseases, currently underrepresented in drug discovery portfolios. Infectious diseases predominantly afflict LMICs, situated mostly in the Global South.

The paper notes, for example, that Africa carries over 95% of the 240 million annual global cases of malaria and 25% of global deaths from TB.

“Historically, efforts to tackle these challenges have principally occurred in the Global North; consequently, African drug discovery efforts have largely been dependent on international funding agencies with programmes driven from abroad,” Chibale explained.

“It is anticipated that lowering these barriers could lead to important scientific contributions from those countries that disproportionately suffer from the bulk of infectious diseases, a milestone towards their eradication.”

Creative Commons License This work is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.

Please view the republishing articles page for more information.

Teaching and Learning


In the news