Kai Ye from Xi’an Jiaotong University in China spoke at the Nanopore Community Meeting in Singapore about “Novel algorithms for long read analysis and applications.” They noted that long reads are “better for structural variant (SV) detection.” However, Ye explained that there are still challenges with long read. They converted the problem from text-based to image-based creating a denoised dotplot. They then use a deep learning approach to detect variants. Ye published the work in Nature Methods last year. There are three components to this framework, and the previous version could only compute one genome at a time. Building on the previous algorithm, now they encode two genomes in one image. For trio data and de novo mutation detection, performance was “promising”very amazing,” according to Ye. The algorithm uses prediction images and algorithms to identify variants. Encoding two genomes on one image and then detecting “smaller” signatures is an innovative system!
