Referee 2.2

[cgreene]

There is a slight imbalance in the presentation of various application areas. The section on drug development (p40ff), for example, is quite extensive and provides a lot of technical details which might be less relevant for a reader who tries to get a general overview of deep learning in biomedical research. An area which is little mentioned on the other hand are deep learning approaches to brain data, eg connectivity maps, and the area of learning from structured data, such as graphs.

[agitter]

Should I cut some of the technical details in the drug discovery section?

Structured data, specifically graphs, seems in-scope to me. We could add a short section about graphs, PPI networks, etc. @zietzm does that still interest you? A graph analysis section could add to #638. We have #241 #543 and perhaps other issues I'm forgetting. This section could be connected to the graph-based methods we discuss in the drug discovery section.

I'm less sure about brain data. There have been many applications of deep learning in neuroscience. A short paragraph may not due them justice, and it is debatable whether it is in the scope of this review.

[zietzm]

Absolutely, I will get on that right away.

[agitter]

@srinituraga agreed to help us with a couple paragraphs on deep learning for neuroscience and connectivity maps!

[cgreene]

Wonderful! :)

[agitter]

The section on drug development (p40ff), for example, is quite extensive and provides a lot of technical details which might be less relevant for a reader who tries to get a general overview of deep learning in biomedical research.

@cgreene I can make revisions to address this comment. Do you think I should trim this section? Remove detail but leave the same references?

[agitter]

#638 added methods for learning on graphs.

[cgreene]

From a read through, the section on ligand-based prediction is one of the longest. However, the section itself on treatments is not that long. What do you think about splitting the representation learning portion at Much of the recent excitement in this domain has come from what could be considered a creative experimentation phase, in which deep learning has offered novel possibilities for feature representation and modeling of chemical compounds into its own subsection?

[agitter]

@cgreene I could split the content, but I'm also wondering whether I would also need to make cuts to shorten the overall length or make it less technical.

[cgreene]

I didn't see anything that stuck out as glaringly obviously overly technical. I found that on a read through it's longer than other areas because it deals with more topics. However, if one considers the part about learning a representation of chemicals as a separate topic, then things are a bit more in balance.

Were there any parts that you saw as too technical? Some of the deeper dives into results touch on important considerations (such as unbalanced training), so I think they're helpful to set things up for later.

[cgreene]

It's worth noting that I may have reviewed this on a previous go around so I may be primed to like it 😉

[agitter]

We have all of the manuscript modifications in place for this comment now. Only the response to reviewers remains.

[agitter]

Closed by #799