Noting that Arjun Krishnan tweeted that he has used a similar approximation (with only the numerator):

Nice work! I've observed that networks created by adding edges between random node pairs with a probability proportional to the product of their degrees in the original network perform really well in many network-based tasks including edge prediction.

Looking forward to seeing the abstract as that helps frame reader expectations.

Also a figure or table would greatly help with connecting to the reader.

Oftentimes a reader will want to look at a table or figure first, since that is quick before deciding to invest the time to understand the prose.

Could the introduction provide a more definitive assessment of when incorporating external information helps models?

One purpose of the review is to frame how scholars think about the topic... not just describing all relevant work.

Take a stand on:

1. the types of ways external info can be incorporated
2. when it is useful to do so.

Paragraphs help readers tremendously. More paragraphs = easier to read and get back on track if lost.

Should this be a paragraph break?

In order to unsterstand a learned classifier, the reader must know three things:

1. what the observations are
2. what the feature/predictors are
3. what the outcome variable is

Make sure that these 3 aspects are clearly established before digging deeper into a specific model.

My understanding is that this review is about using biological knowledge to create bullet 2 (the features/predictors).

Would a table with this information be a good addition to the review?

It may help the reader understand to provide an example of a hand-engineered feature (unless it's too complicated).

The lack of comma after "for images" is confusing. Consider rephrasing like "With images for example,".

I think we need to more strongly quantify that the edge prior can outperform many existing edge predictions to show that there is a major chance of misrepresenting performance if you do not account for it. I will think of phrasings for the abstract after reading whole mansuscript.

edge for consistency

The hetio python package has been renamed to hetnetpy and is now available at https://github.com/hetio/hetnetpy.

This statement is a misunderstanding of Powell 2016, which states:

At Nature, the median review time has grown from 85 days to just above 150 days over the past decade, according to Himmelstein's analysis.

However,

the median review time — the time between submission and acceptance of a paper — has hovered at around 100 days for more than 30 years.

So while the median review time at Nature has gone from 85 to 150 days, this is not the case for all journals. See also the related Tweet.

Just to be clear, the follow up studies contradicted this inference: loss of function reduces LDL levels? If so, I'd be more explicit that the inference turned out to be mistaken.

Typo: a recent

"Project Rephetio" may be a more precise way to refer to our study

Typo: missing a

A good example here is the Network-based in silico drug efficacy screening study. I'd consider adding a paragraph on this study. This study falls into the comprehensive category you mention.

Any idea where this ALL CAPS DOI came from? I've reported this issue with DataCite and it's disheartening to see the issue potentially elsewhere. I'd love to know track down the source of the issue.

should be "diseases"

Remove dash

missing period

Do you mean "any"?

You can hyperlink with the following markdown:

[MedScape](http://emedicine.medscape.com/article/330487-medication)

rather than show the URL

Error

Error

you can add link with the following markdown:

[excel spreadsheet](https://github.com/dhimmel/indications/blob/d3d57b07abcf8d7af55f8e42ef7c4c99acd87ca8/curation/pk/template-pk%20final.xlsx)