Accurate detection of mosaic mutations is the first step in medical research toward developing treatments for many diseases, says Gleeson.
Co-first author and co-author Dr. Xiaoxu Yang, a postdoctoral researcher in Gleeson’s lab, said DeepMosaic was trained on nearly 200,000 simulated biological variants across the genome. Subspecies from data never encountered before.
To train the computer, the authors gave examples of credible mosaic mutations and many normal DNA sequences, and taught the computer to tell the difference. By repeatedly training and retraining with ever more complex datasets and choosing from a large number of models, the computer ultimately identifies mosaic variants far better than the human eye and previous methods. It is now possible. DeepMosaic was also tested on several previously unseen large independent sequencing datasets and outperformed previous approaches.
“DeepMosaic outperforms conventional tools in detecting mosaicism from genomic and exonic sequences,” said co-first author and former undergraduate research assistant at the University of California, San Diego School of Medicine, now a Novartis researcher. Research data scientist Xin Xu said. “The salient visual features detected by deep learning models are very similar to what experts look at when manually examining variants.”
DeepMosaic is free for scientists. This is not a single computer program, but rather allows other researchers to train their own neural networks to achieve more targeted mutation detection using similar image-based setups. It’s an open-source platform, researchers say.
Co-authors include Martin W. Breuss, Danny Antaki, Laurel L. Ball, Changuk Chung, Jiawei Shen, Chen Li, Renee D. George, UC San Diego, and Rady Children’s Institute for Genomic Medicine. Yifan Wang, Taejeong Bae, Alexei Abyzov, Mayo Clinic. Yuhe Cheng, Ludmil B. Alexandrov, Jonathan L. Sebat, University of California, San Diego. Wei Wei, Peking University. and the NIMH Encephalosomatic Mosaic Network.
Funding for this study was provided in part by the National Institutes of Health (grant grants U01MH108898 and R01MH124890), the San Diego Supercomputer Center, and the UC San Diego Institute for Genomic Medicine.
