Approved Research

Transfer Learning Methods for Genomic Prediction to Enable Equitable Understanding of Human Disease

Lay summary

Aims: The proposed project aims to assess the prediction accuracy of transfer learning methods in comparison to standard phenotype prediction methods for various ancestry groups, including populations not of European descent.

Scientific rationale: Precise phenotype prediction is vital for understanding genetic traits and diseases. Traditional methods, like GBLUP and Bayesian Alphabet models, often oversimplify genetic architectures. Machine learning, notably neural networks, offers a more flexible approach to model complex genetic relationships. However, they require extensive data, a challenge for underrepresented populations in genomics studies. Despite genetic variations among populations, shared genetic factors may contribute to traits across diverse groups. This commonality can be leveraged by transfer learning, where knowledge from one model can enhance another working with different data. Transfer learning has shown promise in various genomics tasks, but is in the early stages of research for phenotype prediction. By integrating machine learning, neural networks, and transfer learning into genomics, this research aims to improve phenotype prediction accuracy and enhance our understanding of genetic diversity across populations. This has the potential to advance precision medicine and healthcare outcomes.

Project duration: The intended project duration is 12 months. 

Public health impact: This research can serve as a valuable public health tool by advancing our ability to predict and address diseases more effectively based on genetic diversity. It has the potential to reduce health disparities, enhance personalized healthcare, and inform public health strategies. By focusing on historically underrepresented populations and improving disease prediction, this work can contribute to more equitable healthcare outcomes and better public health interventions.