scRNA-seq filtering thresholds

Question

[melissachua90]

I was given a scRNA seq dataset without any background information (e.g., mitochondrial content). I want to filter n_genes_by_counts.

Before filtering, the violin plot is heavily skewed with a heavy bottom. What cutoff threshold is appropriate?

After I remove n_genes_by_counts > 4000, the violin plot is less skewed but I'm not sure if it is overfitted.