IDBA-UD is a de novo assembler designed for single-cell and metagenomic sequencing data with highly uneven depth. It addresses the challenges of uneven sequencing depths by using a de Bruijn graph approach and incorporating multiple depth-relative thresholds to remove erroneous k-mers. The algorithm also uses local assembly with paired-end information to resolve branch problems in low-depth regions and error correction to improve accuracy in high-depth regions. Compared to existing assemblers like Velvet, Velvet-SC, SOAPdenovo, and Meta-IDBA, IDBA-UD produces longer and more accurate contigs. It performs well on both simulated and real datasets, including single-cell and metagenomic data, achieving higher N50 values and coverage. IDBA-UD iterates k from small to large values, allowing it to reconstruct missing k-mers and reduce errors. It also uses local assembly to handle low-depth short repeat regions and improves scaffolding by aligning reads to contigs. The algorithm is efficient and effective in handling highly uneven sequencing depths, making it suitable for metagenomic and single-cell sequencing applications.IDBA-UD is a de novo assembler designed for single-cell and metagenomic sequencing data with highly uneven depth. It addresses the challenges of uneven sequencing depths by using a de Bruijn graph approach and incorporating multiple depth-relative thresholds to remove erroneous k-mers. The algorithm also uses local assembly with paired-end information to resolve branch problems in low-depth regions and error correction to improve accuracy in high-depth regions. Compared to existing assemblers like Velvet, Velvet-SC, SOAPdenovo, and Meta-IDBA, IDBA-UD produces longer and more accurate contigs. It performs well on both simulated and real datasets, including single-cell and metagenomic data, achieving higher N50 values and coverage. IDBA-UD iterates k from small to large values, allowing it to reconstruct missing k-mers and reduce errors. It also uses local assembly to handle low-depth short repeat regions and improves scaffolding by aligning reads to contigs. The algorithm is efficient and effective in handling highly uneven sequencing depths, making it suitable for metagenomic and single-cell sequencing applications.