As healthcare increasingly moves towards prevention rather than treatment, one of the biggest challenges is identifying who is most at risk of developing disease before symptoms appear. One of the key tools used in genetic testing is the polygenic risk score, which combines information from many genetic variations across a person's DNA to estimate their risk of developing diseases such as colorectal cancer. However, there remains significant room for improvement in how these risk scores are developed and applied.
A National Industry PhD project between The University of Melbourne and Rhythm Biosciences is investigating how advanced machine learning techniques can improve the accuracy and usefulness of polygenic risk prediction for cancer. Using colorectal cancer as a model disease, the research aims to develop new approaches that enhance the predictive power of existing genetic risk models while also addressing a major limitation in the field: the reduced performance of many risk scores when applied to people from non-European ancestry backgrounds.
The real-world problem is both scientific and practical. Most existing polygenic risk scores have been developed using data from people of European ancestry. As a result, they do not always perform equally well across diverse populations. At the same time, even in populations where they perform well, opportunities exist to improve their predictive accuracy. These limitations matter because accurate risk prediction is essential for delivering personalised healthcare. If risk can be assessed more precisely, healthcare providers can better identify individuals who may benefit from earlier screening, closer monitoring or preventive interventions. Conversely, people at lower risk may be able to make more informed decisions about the intensity and frequency of future screening.
To address these challenges, the project combines statistical genomics, epidemiology and machine learning. The research is investigating whether sophisticated machine learning approaches can improve the predictive performance of traditional polygenic risk scores and whether new methodologies can adapt those scores for use across different ancestry groups. Using large-scale datasets including the UK Biobank, the project is examining how genetic variation, disease prevalence and biological pathways can be incorporated into more powerful prediction models.
The potential benefits are substantial. More accurate risk prediction could support earlier detection of colorectal cancer and potentially improve prevention strategies. The methodologies developed through the project may also be transferable to other cancers and diseases, creating opportunities to improve personalised healthcare across multiple clinical areas. The research supports a broader shift towards precision medicine, where prevention, screening and treatment decisions can be tailored more appropriately to an individual's genetic risk profile.
For PhD candidate Max Schuran, the project provides the opportunity to work at the intersection of machine learning, genomics and population health. Max has developed expertise in state-of-the-art polygenic risk score modelling, machine learning, deep learning, epidemiology and genetic analysis, while also gaining experience working with large-scale genomic datasets and cloud-based research environments. The project has supported participation in specialist training programs, machine learning communities, genomics courses and international academic forums, helping build a highly sought-after combination of scientific and technical skills.
Industry partner Rhythm Biosciences benefits equally from the collaboration. The project directly addresses challenges relevant to the development of risk prediction products and aligns closely with the organisation's strategic focus on genomic risk assessment. Findings generated through the research are intended to contribute to future improvements in risk prediction tools, supporting the translation of cutting-edge research into products that can be used in real-world healthcare settings. Access to new methodologies, advanced machine learning approaches and rigorous evaluation frameworks strengthens the organisation's ability to continue innovating in an increasingly competitive personalised health market.
The collaboration is also creating valuable scientific outputs. During the project, novel evaluation frameworks have been developed, manuscripts have been submitted and accepted for publication, and benchmarking studies have been undertaken comparing machine learning-based approaches with existing state-of-the-art models.
The significance of this research extends beyond colorectal cancer. The methodologies being developed could improve how genetic risk is assessed across a wide range of diseases, helping unlock more accurate and equitable approaches to personalised healthcare. For industry, the project offers a pathway to more powerful risk prediction tools that can be incorporated into future products and services. For patients and healthcare providers, it represents another step towards earlier detection, targeted prevention and more informed health decisions based on an individual's unique genetic profile.