RAVE:Tutorial Outline

From OpenWetWare
Jump to navigationJump to search

Flow for analysis


1. Get the raw data imported A. Directory tree i. iEEG data ii. Freesurfer data iii. Others B. File format C. Collect necessary parameters (sampling rate)

2. Pre-processing A. Notch B. Wavelet C. Epoch D. Localization

3. Single Subject Analysis. Goal: Identify the electrodes and conditions that should be passed on to group-level analyses

Save analysis parameters

Repeat 1-2 with new subjects. For step 3, can Load settings from previous subject, but be sure to inspect what you're seeing. If you make any changes, re-save the settings file (but think about exploratory / confirmatory analysis issues).


4 (Optional). Clustering Goal: If you suspect heterogeneity of response across electrodes but you lack a clear anatomical separator, then you can cluster the electrodes (across subejcts) to determine functional groups. Once you have determined these functional groups, you can use this derived variable in your group-level confirmatory analysis. Beware of circularity here. If you cluster electrodes into groups based on, e.g., their response to Auditory only stimuli, it is not meaningful to test whether the resulting have a "significant difference" (p < 0.05) response to auditory stimuli.

A. Load up the data B. Choose clustering parameters i. distance metric ii. agglomeration method (e.g., Ward's) iii. Desired number of clusters C. Run algorithm D. Export data i. This will be in FST format (dramatic improvement in size and write/read speed)

4 / 5. Group-level analysis: LME Goal: Are the differences between conditions consistent across subjects? A. Load up the exported data from each subject (if you are using the exported cluster data, this will be a single file) B. Build your own ROI if needed i. Choose variable ii. Create ROI groupings iii. Assign levels of the ROI variable to each grouping C. Create ConditionGroup (auto-populates from .yaml) i. Create condition groupings ii. Assign conditions (levels of the Condition variable) to each grouping

D. Build your LME Model i. Pick you depdendent variable (if you have 1 event, there will be only one option) ii. Decide how you'll use your ROI a. Pick variable (CustomROI if you defined one in B) b. Pick Analysis type - Filter only: inclusion/exclusion based on levels selected, no fixed effect - Stratify: included as interaction term in the LME , but post-hocs are done w/n ROI group) - Full interaction: included as interaction term in the LME, post-hocs allow for comparison across ROIs - Difference of Differences: included as interaction term in the LME. Post-hocs compare the size of the pairwise differences between each level of the Fixed Effect across ROIs. e.g., calculate AUD-VIS for each ROI, then compare these deltas across ROIs. iii. Pick your Fixed Effects iv. Pick your Random Effects v. Run it!