This is a conceptual question regarding ALE implementation in NiMARE that I wanted to clarify. Are the activation results based on all contrasts (positive and negative) from the referenced studies or just the positive contrasts? If the results contain both positive and negative, is there then a way to separate them?
Thanks for the clarification.
ALE essentially just compares the density of peaks between sets of studies, so you don’t want to mix positively- and negatively-valued peaks (e.g., A > B and B > A) in a single ALE analysis. You can use an ALESubtraction to compare A > B and B > A contrasts across studies.
Typically, papers would either separate the A > B and B > A peaks into different tables or would include the test statistic (e.g., z, t) associated with each peak, which would have the direction included (e.g., z = 1.65 for A > B and z = -1.65 for B > A).
If you’re using a large, automated dataset like Neurosynth, then the short answer is no. The general hope when using a dataset like Neurosynth is that the signal from such a large number of studies will overwhelm any noise from things like badly extracted coordinates, mixed positive and negative peaks, and coordinates extracted from multiple analyses within a given paper. You could always hunt down the papers and correct the coordinates/contrasts, but that would be quite the endeavour. Neurosynth Compose is working on making that a reasonable procedure though: https://compose.neurosynth.org/
Thanks @tsalo!
Just to be sure I’m understanding correctly, it’s not really possible to differentiate positive/negative peaks with NeuroSynth (which I’m indeed using), but ideally studies would properly indicate this information, thereby not contributing to noise?
Is the ALESubtraction class meant to be used in lieu of the nimare.meta.cbma.ale.ALE() and nimare.correct.FWECorrector() steps, or once those steps have completed?
Correct, and some day, with user contribution, Neurosynth Compose will have that level of detail. At the moment, however, it’s just not available.
It’s a separate Estimator, so you use it instead of the ALE and FWECorrector tools. It’s not compatible with any of the multiple comparisons correction methods in NiMARE, unfortunately, but it uses random group reassignments to build a null distribution. Here’s an example in our documentation that covers using it: NiMARE: Neuroimaging Meta-Analysis Research Environment — NiMARE 0.0.13+0.g9f6ea22.dirty documentation
Got it, I’ll give the ALESubtraction() a try, thanks again
MANAGED BY INCF