I appreciate the detailed response! I meant to come across as more curious than critical but in any case you have satisfied my curiosity. I also admire your forthcoming-ness.

Keep up the great work!

In this first iteration we didn’t do any NLP or semantic parsing, so we can’t distinguish positive from negative findings.

You can read the preprint version of our manuscript with the full methods on the site, here:
http://www.brainscanr.com/Paper

We try to be as forthcoming about the limitations as possible. Regarding this point you raise, we say:

“Of course, there are limitations to the method we used to create these association networks. Most striking is that these relationships are inherently based upon the existing literature, and thus the observed associations may reflect publication biases rather than necessarily true associations. Also, our method does not differentiate negative from positive results such that, if a paper’s title or abstract states that the amygdala does not relate to fear, that paper is weighted as strongly as a paper that finds a positive relationship between the amygdala and fear. Nevertheless, there is a well-described publication bias in the literature such that negative results are under-reported, and thus the relationships we discover are more likely to be biased toward positive associations, though that bias cannot be quantified.”

Or, as we say (more flippantly) on the site:

“Luckily, there is a positive publication bias in the peer-reviewed biomedical sciences that we can leverage to our benefit (hooray biases)!”

I wonder if/how the map designers accounted for words co-occurring in studies showing that they were not correlated. Can anyone find a strong connection between two concepts that should be negatively correlated?

Not sure if you saw this, but we can also create and entire connectivity diagram for the inferred associations between brain regions. Makes for a really nice graphic:

http://www.brainscanr.com/Paper

We’re continually tweaking the site, and we will be updating it regularly to add more terms to the database.

Because of the access limitations set by PubMed, we cannot access data on the fly, thus the current version of our site only works for our pre-populated database of terms. Thus if a term is *not* in the database, it cannot be searched for, nor will it show up as related to any other terms.

We’re working on a way to calculate these relationships on the fly, but for now, we’re a bit limited.

The relational database grows quickly (as the square of the number of terms, since every term must be associated will all the others in the database), so the more terms we add, the harder everything becomes.