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

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Existing genome assembly frameworks struggle with low-coverage or error-prone sequencing data, failing to maintain genome integrity and biological variation.
• Challenges include path entanglement in reconstructing short sequences and reduced accuracy under noisy conditions.

Purpose of the Study:

• To develop an advanced assembly graph framework that effectively handles low-coverage and noisy sequencing data.
• To improve genome integrity, biological variation preservation, and data reconstruction accuracy.

Main Methods:

• Introduction of a dynamic variable-order unitig-level assembly graph (DVOUG).
• DVOUG constructs an initial precise unitig graph with a high k-value, adaptively lowering it in low-coverage or noisy regions.
• Utilizes graph neural networks (GNNs) for edge prediction within the DVOUG framework.

Main Results:

• DVOUG successfully addresses path entanglement issues in low-coverage scenarios.
• Demonstrates superior performance in genome assembly and DNA storage data reconstruction compared to previous methods, even under low coverage.
• Achieves over 99% recall rate for edge prediction using GNNs, outperforming traditional methods and reducing training time by 4×.

Conclusions:

• DVOUG is highly effective in handling complex, noisy sequencing data.
• Enhances assembly accuracy, connectivity, and learnability, showing significant potential for practical genomic applications.
• Represents a substantial advancement in bioinformatics for challenging sequencing conditions.