Machine Learning for Interpreting Rare Variation in Comprehensive Newborn Screening and Pharmacogenetics

In California 500,000 babies are born each year, some of whom have genetic mutations that cause disease or altered responses to medications. Recognizing which genetic variants cause problems is surprisingly difficult–harder than finding a needle in a haystack, where once you find the needle, you know it’s different from the hay.  Today, when studying differences in DNA sequences, we can’t tell which will cause disease. So, it’s more like finding a “poisoned needle” in a “needlestack.” We are developing new methods to identify “poisoned needles,” the damaging variants in DNA.  We will search for the mechanisms by which variants impact the function of genes.  With experts in biology, computer science, medicine, and ethics from Stanford, UCSF, and Berkeley this project, funded by the Chan Zuckerberg Biohub, will focus on serious newborn diseases and on gene variants that require customized drug choice and dosage. Studying ethical dimensions of data use will bring together bioethics scholars from the three institutions at the Biohub.

Machine Learning for Interpreting Rare Variation in Comprehensive Newborn Screening and Pharmacogenetics