Abstract
While de novo variants cause many Mendelian disorders, their detection currently requires sequencing of the proband and both biological parents. This is not feasible when only one parent is available, a limitation for millions of families. We developed duoNovo, which identifies de novo variants from single parent-proband duos using long-read sequencing followed by haplotype reconstruction and detection of identical-by-descent haplotype blocks. We sequenced 40 trios and applied duoNovo to each of the 80 duos constructed by masking one parent, classifying over 20 million variants. We evaluated duoNovo's performance against classifications obtained using the full trios (which included over 1,900 de novo variants), demonstrating very high precision and low error rate, and perfect accuracy among variants absent from gnomAD. duoNovo is freely available to the community as an R package, and may represent an example where long-read sequencing provides clear diagnostic benefit over short-read sequencing.