AI-Assisted Genome Studies Are Riddled with Errors

November 13, 2024

3D rendering of a DNA molecule

(The Scientist) – Researchers used artificial intelligence in large genomics studies to fill in gaps in patient information and improve predictions, but new research uncovers false positives and misleading correlations.

With AI-assisted GWAS [genome-wide association studies], Lu and his colleagues noticed false associations between gene variants and type II diabetes. For example, four gene variants showed a high correlation with the disease in an AI-assisted GWAS, but not when using a conventional GWAS approach. However, previous research has shown that although these genes act on a cellular pathway that is indirectly connected to blood glucose levels it does not strongly influence them. (Read More)