Settings for Neurosynth.org

[Ernst Schwartz]

Hi!

I have 2 questions regarding the neurosynth.org website vs. api...

1) on the web, the dictionary contains roughly 1300 terms , in the last data release on github there's more than 3000. How where to 1300 selected and is there a way of obtaining the list?

2) how are the r2 values calculated in the web interface? I'm getting different (lower) results using the decoder (pearson correlation) with the latest (2018) Dataset 

if the neurosynth web interface is too different from the latest release, is there any way of querying it programmatically remotely (neurosynth-web mentions 300GB of data ... too much for my machines :) and how would one cite the results?

and thanks a lot for all you effort and providing such a wonderful tool!

best

Ernst

[Taylor Salo]

> 1) on the web, the dictionary contains roughly 1300 terms , in the last data release on github there's more than 3000. How where to 1300 selected and is there a way of obtaining the list?

I'm not really familiar with the website code, but my guess is that this difference can be explained by the stop lists in the neurosynth-web repository (https://github.com/neurosynth/neurosynth-web/tree/a15f3fc81b0bdb7bb0c643c48b34a9e516f46b9f/data/assets/misc). You can try taking the diff between the terms in the Neurosynth dataset and the stop lists to see if the resulting list of terms matches up with the website.

> 2) how are the r2 values calculated in the web interface? I'm getting different (lower) results using the decoder (pearson correlation) with the latest (2018) Dataset 

At a glance, it looks like the decoders operate differently, although it depends on the scaling for the meta-analytic maps on the website.

Here's the website decoding code: https://github.com/neurosynth/neurosynth-web/blob/a15f3fc81b0bdb7bb0c643c48b34a9e516f46b9f/nsweb/tasks/__init__.py#L150-L155

Here's the package decoding code: https://github.com/neurosynth/neurosynth/blob/9022f1b5a9713dedc9d8836ed023aaba2a983d87/neurosynth/analysis/decode.py#L240-L243

I don't know if there's a way to access the maps programmatically, but web-scraping is explicitly warned against in the FAQ.