Researchers from the University of California San Diego School of Medicine and Rady Children’s Institute have developed a deep-learning model called DeepMosaic that can detect mosaic mutations in DNA faster than human medical geneticists. Mosaic mutations are DNA mutations present in a small percentage of cells within normal DNA sequences and are currently difficult for most DNA mutation software detectors to identify. The process of reviewing DNA sequences by eye can also be time-consuming and prone to error.
How DeepMosaic Works
Deep learning, a type of machine learning that uses artificial neural networks to process visually represented data, has shown high potential in learning from large amounts of data and performing necessary tasks. To train DeepMosaic, the researchers provided it with examples of mosaic mutations and normal DNA sequences and used increasingly complex datasets to enable it to identify mosaic mutations better than humans and previous methods. The model was then tested on independent large-scale sequencing datasets and found to be more effective at identifying mosaic mutations than previous methods and humans.
Importance of Detecting Mosaic Mutations
According to senior study author Joseph Gleeson, detecting mosaic mutations is the first step in developing treatments for various disorders, including epilepsy. Mosaic mutations within the brain can cause epileptic focus in epilepsy patients who have not responded to common medication and require surgical excision of the short-circuited focal part of the brain to stop seizures. DeepMosaic has allowed for the improvement of DNA sequencing sensitivity in certain forms of epilepsy and led to the discovery of new ways to treat brain disease.
Use of Deep Learning in Clinical Care
Deep learning is being increasingly applied to clinical care. In November, researchers from Massachusetts General Hospital and Brigham and Women’s Hospital developed a deep-learning model that can predict the 10-year risk of death from a heart attack or stroke by analyzing X-ray images. In August, a deep-learning tool was found to help neuroradiologists diagnose brain tumors using MRI scan data. The tool was trained to evaluate images and characterize various types of intracranial tumors, achieving high accuracy, sensitivity, and specificity when tested on several datasets.