In a fixed ridge test in three mice, traced sister cells cut mean calcium prediction error by 23.0% for excluded odor modules

A common anatomical input can guide prediction of a neural response. We tested whether traced sister cells in the mouse olfactory bulb help predict target calcium responses to odors excluded from fitting. We used 370 cells in 66 glomeruli from three anesthetized mice. Each test excluded a complete odor delivery module and used disjoint source and target repeat halves. A fixed ridge model added the mean sister response after removal of its training fit to a plane response. This cut error by 30.33%, 37.79%, and 0.94% in the three mice. The equal-mouse mean decrease was 23.02%. The decrease against source-count and plane-matched nonsister sets was 23.11%. A strict-cell check with nearest nonsisters gave a mean decrease of 23.98%. Thus, traced input groups carried response information that transferred across the tested odor modules. Source cells still observed the test odors; target cells supplied calibration data for other odors. The result concerns slow calcium signals and does not identify direct synapses or a circuit mechanism. The small third-mouse effect and the sample of three mice limit general claims. The package preserves compact inputs, source provenance, code, every main fold loss, source sets, figures, fixed checks, and repeat instructions.

olfactory bulb · shared input · anatomical tracing · calcium imaging · response transfer · cross validation