Mosaic mutations are successfully identified via deep-learning technology

Mosaic mutations are successfully identified via deep-learning technology

Genetic mutations can lead to various disorders that are currently not fully understood or able to be treated. One type of mutation, called mosaic mutations, which occur in a small percentage of cells, can be challenging to detect because they are present in only a small number of cells.

While scanning the 3 billion bases of the human genome, the current DNA mutation software detectors are not well equipped to identify mosaic mutations that are mixed in with normal DNA sequences. Medical geneticists frequently have to examine DNA sequences by hand in order to try to detect or confirm mosaic mutations—a time-consuming process rife with the potential for error.

Researchers from the Rady Children’s Institute for Genomic Medicine and the University of California San Diego School of Medicine describe a method for instructing a computer how to recognize mosaic mutations using an artificial intelligence technique known as “deep learning” in a paper that will appear in the January 2, 2023 issue of Nature Biotechnology.

Deep learning, also known as artificial neural networks, is a machine learning technique that trains computers to learn from examples, particularly from a lot of information, just like humans do. In contrast to conventional statistical models, deep learning models process visually represented data using artificial neural networks. The models work in ways that are comparable to how humans absorb visual information, but with far more precision and attention to detail. This has greatly improved computing capabilities, including mutation detection. Focal epilepsy is one example of an unsolved condition, according to senior study author Dr. Joseph Gleeson, Rady Professor of Neuroscience at the UC San Diego School of Medicine and head of the neuroscience research division at the Rady Children’s Institute for Genomic Medicine.

Four percent of people in the population have epilepsy, and roughly one-fourth of focal seizures don’t respond to standard treatment. To stop seizures in these patients, the short-circuited focal area of the brain is frequently removed surgically. Mosaic mutations in the brain of these patients can result in epileptic focus.

When we used our ‘DeepMosaic’ technology to analyze the genomic data, the mutation in many of the epilepsy patients that we were unable to identify the source of became clear. This has made it possible to increase the sensitivity of DNA sequencing in some types of epilepsy and has produced findings that suggest novel approaches to treating brain diseases.

According to Gleeson, the first step in medical research toward creating cures for many diseases is the correct diagnosis of mosaic mutations.

DeepMosaic was trained on nearly 200,000 biological and simulated variants across the genome, according to co-first and co-corresponding author Xiaoxu Yang, Ph.D., a postdoctoral researcher in Gleeson’s lab, until “finally, we were satisfied with its ability to detect variants from data it had never encountered before.”

The authors supplied the computer examples of reliable mosaic mutations as well as several typical DNA sequences to train the machine to distinguish between the two. The computer was eventually able to recognize mosaic mutations considerably better than human vision and earlier methods by constantly training and retraining with ever-more complicated datasets and selecting amongst a dozen models.

DeepMosaic, which was developed by Xin Xu, a co-first author and former undergraduate research assistant at the UC San Diego School of Medicine who is currently a research data scientist at Novartis, outperformed conventional techniques in detecting mosaicism from genomic and exonic sequences. The deep learning models’ top visual attributes are extremely comparable to what experts look for when personally evaluating variants.

Scientists get free access to DeepMosaic. The platform, which is open-source, enables other researchers to train their own neural networks to achieve a more focused detection of mutations using a similar image-based configuration, the researchers said. It is not a single computer program.

Reference: Yang X, Xu X, Breuss MW, et al. Control-independent mosaic single nucleotide variant detection with DeepMosaic. Nat Biotechnol. 2023:1-8.
DOI: 10.1038/s41587-022-01559-w

Also read : Here are ten potential New Year’s resolution ideas 2023

Leave a Comment

Your email address will not be published. Required fields are marked *