Duty Calls, Science

If I don’t get it, you should be concerned.


The latest post at Shit My Reviewers Say is “My first concern is that I don’t get it.

And the obvious response is illustrated by a picture saying “Your problems with me are not my problems, those are your problems.

What can you do as an author if the reviewer is just too stupid to understand your ingenuity?

But … and this is a big but … there are areas of research where I would use that reviewer’s comment myself. If, say, you are writing about probabilistic models in cancer genomics and I can’t make any sense of what you are saying, it is your problem, not mine.

Here is an example. The ABSOLUTE paper on “Absolute quantification of somatic DNA alterations in human cancer”.

The paper itself is fine. It’s all sugar and lollipops. The actual meat is in the supplementary method description, which fewer people might have read in depth.

The core model of ABSOLUTE is a Gaussian mixture model, where the mean of each component is given by the absolute copy number mapped into the measured (relative) scale using purity and ploidy of the sample. You can find it as Equation 4 in the Model section of the paper.

That part of the model is pretty straightforward. What makes it work well is the choice of the multinomial distribution generating the mixture weights.

The authors must have foreseen trouble because they begin the next paragraph with the words “Some complication arises …” — and indeed it does. Maximum entropy stuff falls from the sky and I have no clue what’s really happening.

And in case you are thinking I’m just being stupid again: I tried it on the participants of the Simons Institute workshop on Algorithms in Genomics in Berkeley earlier this year and the only answers I got were things like “Yeah, what’s going on there? No one understands that bit.”

If you treat the method as a black box and don’t care how it works, it might not matter to you how well it is described. Fine. But for a methods person such lack of clarity is really worrying. How can I judge why this method works well and how it is related to other approaches if I don’t even understand how such a central bit of the model works.

My first concern is that I don’t get it.

My second concern is that no one else does either.

Florian

Advertisements

You gotta talk to me!

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s