Machine Learning to Develop Risk Score for Glaucoma
Glaucoma can be prevented through timely diagnosis and treatment, but most at-risk individuals, like Jack, are unaware of their current/future problems.
Identifying the underlying susceptibility for glaucoma in asymptomatic (healthy) populations can identify lower-risk individuals for less intensive monitoring and higher-risk individuals for aggressive, prospective monitoring.
Such risk stratification is accomplished through the study of thousands of environmental, physical, and biological parameters (including genetics) to identify those that best predict future disease. For example, intraocular pressure (a modifiable risk factor in glaucoma) in the healthy population is influenced by several genetic variants (i.e. intraocular pressure is a polygenic trait). However, these genetic variants each contribute just a very small amount to the total risk.
Therefore, predicting glaucoma risk from so many features is fundamentally and technically challenging (combinatorically enormous number of ways that a genome can be altered). To accomplish this goal, our research group utilises large, multi-cultural, international datasets and machine learning to predict the environmental and biological features which best predict risk for glaucoma (onset) and further disease progression (need for surgical intervention). These computational approaches provide functional interpretation of the impact of genetic variation to identify personalised risk factors.
In the future, the prevalence of test-at-home DNA kits, such as those sold by 23 and me, will make personalised genetic diagnoses common, leveraging well-functioning risk-scoring algorithms to diagnose future glaucoma risk and identify those in need of prospective monitoring and treatment.
EYE Believe… that well-functioning risk scoring algorithms can leverage our genetics to better predict glaucoma onset, preventing vision loss in thousands of kiwis.