Associate Professor Qiongshi Lu and colleagues were highlighted in a Nature Genetics article for their work on the perils of relying on AI to predict complex traits and disease risks with genome-wide association study (GWAS) data, which can have limited data the AI attempts to fill in.
The study shows that a common type of machine-learning algorithm employed in genome-wide association studies can mistakenly link several genetic variations with an individual’s risk for developing Type 2 diabetes. These algorithms can produce false positives when the prediction is based on correlating all genetic variations as actual diabetes, even though this is not the case.
This problem is pervasive in AI-assisted studies. Dr. Lu and team developed a new statistical method for guaranteeing the reliability of AI-assisted GWAS. The method helps reduce the bias that can be introduced when using machine learning algorithms.
These studies highlight the importance of statistical rigor in AI-assisted and biobank-scale research studies.