Reviews: A coupled autoencoder approach for multi-modal analysis of cell types

Originality: Multimodal data is increasingly becoming available in various omics field. Notably in neuroscience, patch-seq has been recently developed to profile neurons both transcriptomically and electrophysiologically (Cadwell et al, 2016, Fuzik et al 2016). Now, the first large data sets are becoming available, yet analysis methods that can fully leverage the multimodal data sets are still largely missing (see Tripathy et al, 2018; Tripathy et al. 2017, Kobak et al. 2018). The present submission extends prior work in coupled autoencoder architecture to patch-seq and equips them with a new loss function for the coupling loss that does not allow for degenerate solutions. Quality: Overall the paper appears to be well done – it almost contains a bit too much material for such a short format.