DVOUG enables robust DNA sequence assembly and reconstruction with a dynamic, variable-order graph

Under low-coverage or error-prone sequencing conditions, existing assembly frameworks often fail to simultaneously preserve genome integrity and biological variation. To address these, this work introduces a dynamic variable-order unitig-level assembly graph (DVOUG), which constructs an initial precise unitig-level assembly graph using a high k-value and progressively lowers the k-value in regions with low coverage or high noise. Experimental results show that DVOUG solves the problem of path entanglement when reconstructing short sequences under low coverage and significantly outperforms previous graphs in both genome assembly and DNA storage data reconstruction tasks, even under low coverage. In addition, DVOUG achieves more than 99% recall rate by graph neural networks (GNNs) for edge prediction, exceeding both unitig-level assembly graphs and traditional DBGs, while also reducing training time by 4×. In summary, DVOUG excels in handling complex noisy data, enhancing assembly accuracy, connectivity, and learnability, with strong potential for practical applications.