Embedding quality metrics of different methods on simulated data sets with various dropout rates. Q values measure the local (A) and global (B) structure preservations. Larger values mean better preservations. Each setting generated 10 data sets.