With 0.5 s filters and ridge peer controls, a selected unit gave an uncertain 7.4% calcium prediction gain in 20 rat spines

A functional map can link a spike source to a dendritic spine, but shared activity can limit source identification. We tested a selected source after the other recorded units were given. We used all 20 exported spine records from 19 distinct recordings in 10 rat cortical cultures in DANDI 000223 version 0.260528.0906. Each recording half selected and fitted a candidate for testing on the other half. With 0.5 s filters, a nested ridge peer model had mean normalized test error 0.42445. Adding the selected past unit with an unpenalized nonnegative coefficient reduced error to 0.39301, a 7.41% decrease. The 95% whole-culture bootstrap range was -2.64% to 18.79%. Six cultures improved. A future trace gave a 0.16% decrease. Filter and penalty checks changed the added-unit result. An exact duplicate-source simulation gave a mean 16.3% prediction gain despite adding no new linear predictor direction. A simple proof shows why: an unpenalized candidate can change the ridge penalty without adding information. Thus, the rat sample did not establish a general added prediction gain, and the score cannot by itself identify unique input information. We corrected unstable peer feature scaling after the first run and retain that run and the amendment. Independent fits agreed within 4.0e-12, a new extraction matched 209 source arrays, and an offline repeat matched all result tables and prediction arrays. The package preserves data, provenance, code, initial and corrected results, figures, diagnostics, proof, and repeat instructions. These offline tests do not establish physical synapses or reject source-specific transmission.

dendritic spine · functional connectivity · spike source · calcium imaging · ridge regression · source identification · prediction · rat culture · reproducible analysis