Leveraging Machine Learning for Precision Medicine In Ghana

Abstract

Precision medicine, empowered by machine learning, has made substantial strides in developed countries, yet its implementation in Ghana faces substantial challenges due to limited research and data on African populations. This thesis addresses the critical need to establish systems for delivering personalized treatments using machine learning in Ghana’s healthcare context. Hence, it investigates the question, “can machine learning be leveraged to develop personalised plans in the context of Ghana’s personalized healthcare system?” This study combines Node2Vec graph embeddings with cosine similarity calculations to predict novel drug-disease association, focusing on drug repurposing as a cost effective strategy for personalized medicine. The study employs a two step prorcess, first, applying machine learning techniques to the well-established PrimeKG dataset, then comparing these approaches with a database contextually representative of Ghana. Experiments are conducted on both foreign data (focusing on bupropion) and Ghanaian data (focusing on AMG-510 or Sotorasib), utilising molecular profiling data from breast and colorectal cancer patients. Results demonstrate the potential of the approaches in identifying drug repurposing opportunities relevant to the Ghanaian population. The foreign data experiment reveals new potential indications and contraindications for bupropion in cancer related fatigue and in seizures respectively, whilst the Ghanaian data experiment uncovers promising associations between AMG-510 and NRAS-related cancers, particularly gastric adenocarcinoma. These findings provide a foundation for leveraging machine learning in Ghana’s healthcare system, potentially reducing drug development costs and improving access to personalized treatments.While challenges remain, this study marks a significant step towards implementing precision medicine in Ghana.