Over polishing genome assemblies after using Neural network based polishers

Question

[urmi208]

Hi all,

I have noticed that after running RNN (Recurrent-Neural-Network) based polishing programs like HELEN/medaka, the quality of the assembly worsens in terms of BUSCO completeness, N50, etc. which suggests some over-polishing. I am a bit puzzled why would it do that. Does anyone have any opinions on that?

Cheers, Urmi