N50 become smaller while increasing the reads depth about arcs HOT 2 CLOSED

no, nothing wrong. With increased read depth, the scaffold graph increases in complexity, and some merges that were unambiguous at lower read coverage may no longer be possible under the same parameter combination. Although, judging from the N25 length you reported, there are still gains to be had with increased read depth for large scaffolds comprising 25% of your assembly bases. I'd say this is a great result, you have effectively tripled your N50 length! You may also want to report the number of sequences >500bp (or 1kbp) / Number of merges with increased coverage and if you plot depth vs. cumul merges, my intuition is that the number of merges will start to plateau.

It might be interesting to separate your dataset into three sets of ~50X each, and perform 3 iterative scaffolding rounds, with the same parameters. Because barcodes are re-used, the more data one uses at once with ARCS/LINKS may have unintended consequences, and confound the scaffolding process.

Thank you for sharing your results with us,
Rene

from arcs.

@warrenlr Thanks for your reply, I'll have a try~

from arcs.