With 0.5 s filters, a past spike rule reduced calcium test error by 30% after shaft and population controls in 20 spines from 10 rat cultures

We tested whether a spike source selected in one time block predicts a spine calcium trace in another block. We used all 20 exported spines in 19 distinct recordings from 10 rat cortical cultures in DANDI 000223 version 0.260528.0906. A fixed linear rule used 0.5 s exponential spike filters. The baseline included adjacent shaft, time, and past and future mean population traces. Each test block was excluded from unit selection and coefficient fitting. Mean normalized test error fell from 0.63719 to 0.44317, a 30.45% decrease. The 95% whole-culture bootstrap range was 5.10% to 48.05%. Eight cultures improved. Removing the culture with the largest contribution left an 11.87% gain. A future spike rule gave a 5.11% gain, with a range from -3.29% to 9.62%. The past-minus-future gain was 25.34 percentage points, but its two-sided culture sign-flip p value was 0.0723. The same past unit was selected in both halves for 8 of 20 spines. Thus, added calcium prediction did not ensure stable input identity. Planned filter and low-activity checks kept positive mean gains. Independent fits and a second source extraction matched the results. A range reader reduced the 84.3 GB source to a small reproducible test package. These are offline signal tests in selected cultured spines. They do not establish physical synapses, conductance or a causal route. Shared activity, ten culture IDs and unknown donor overlap limit the inference. We preserve source arrays, provenance, code, predictions, all results, figures and repeat instructions.

functional connectivity · dendritic spine · calcium imaging · spike trains · time block validation · culture · input identity · DANDI · reproducibility