Synapse type data increased test error by 5.6% in fixed ridge tests that excluded both regions in one 35-region mouse atlas

We tested whether regional synapse type data improve predictions of awake fMRI coupling beyond a structural connectome and spatial position. The data had 35 right hemisphere regions and ten fMRI mice. Each test excluded all response pairs that touched either test region. At ridge penalty one, adding three synapse density maps increased mean squared error from 0.0257395 to 0.0271773, or 5.59 percent. Random pair tests allowed regions to occur in training and test data. They gave error decreases of 7.97 to 9.55 percent. Matching the training pair count kept this gain. An added test selected penalties within the training data and gave a 4.77 percent error increase. The 34 pairs that touched RT contributed more excess error than the net increase. RT was the only thalamic region and had the lowest density for all three types. Removing its pairs from the aggregate score gave a 4.55 percent decrease. An added density range check marked 99 pairs with higher error. The central range from 999 mouse resamples was -15.50 to +3.59 percent for the error decrease. The results concern one atlas, fixed anatomical maps and the tested pair models. They do not establish a lack of functional roles for synapse types. The package has source data, checked atlas mappings, all predictions, code, figures and repeat instructions.

connectome · synaptome · structure and function · functional connectivity · fMRI · mouse brain · cross-validation · ridge regression · extrapolation · PSD95 · SAP102