With each mouse excluded from ridge fitting, visual similarity reduced mean squared log voltage error by 3.64% in 92 mouse V1 connections
We tested whether visual response similarity helps predict synaptic voltage magnitude after measured cell properties are supplied. We used version 1 of the public Znamenskiy mouse V1 data. We rebuilt the original sample of 52 PV-to-pyramidal and 40 pyramidal-to-PV connections in 20 mice. Each outer ridge fit excluded all connections from its test mouse. Inner tests chose the penalty. Controls were soma distance, source and target depth, and baseline target voltage. Adding similarity reduced mean squared log10 voltage error from 0.150493 to 0.145021, or 3.64%, with equal direction weights. A descriptive 95% mouse-resample range was -7.10% to +13.22%. A later 1,000-draw check repeated all inner and outer fits. Its range was -18.46% to +30.41%. Both direction ranges then included zero. This gives no clear combined gain and does not establish equal performance. Inhibitory error fell by 16.40%; excitatory error rose by 2.58%. Simpler controls gave larger combined gains. The original counts and correlations were reproduced. Independent calculations checked 628 response amplitudes and 704 outer fits. A full source rebuild and repeat reproduced all 14 numerical files exactly. The package preserves compact data, code, models, predictions, figures, source hashes, and repeat instructions. These are local physiological connections. Baseline voltage controls do not measure conductance. The ranges are descriptive and have no exact coverage guarantee for a new experiment.
visual cortex · synaptic weight · PV interneuron · paired recording · visual response similarity · nested cross validation · reproducible benchmark
References
6- doi:10.1016/j.neuron.2023.12.013
[Legacy relation: reproduces] We reproduce the main connection counts and visual similarity correlations, then test added prediction value with complete mice excluded from fitting.
- usesdoi:10.25418/crick.23295188.v1
Pinned source data and method definitions. We use connections.mat and pv_rois_vivo_vitro.mat from Data.zip file 41074013.
- usesWith test mice excluded from model fitting, reverse excitation gave no clear added strength prediction gain after same-PV calibration in 53 mouse V1 pairs sci:sha256:4056880473e69ea8211e302707dc781052484386e1d388b523040c3d418a6bdc
We reuse its checked cache of the unchanged raw patch source. Its prediction task uses reverse excitation; ours adds matched visual data and tests visual similarity. No prior analysis code or derived values are reused.
- Paired recordings in mouse V1 found most excitatory but less than half of the inhibitory connections that MICrONS electron microscopy predicted for 17 matched cell classes sci:sha256:ce046109af65d9a43f99d098f26cd85017dc112137362e3deed08cf3846aea5f
Prior SCI work on the relation between structural connection presence and detected physiological responses in mouse V1. It uses other data.
- usesdoi:10.1080/00401706.1970.10488634
Established ridge regression method.
- usesdoi:10.1186/1471-2105-7-91
Established separation of inner model selection from outer prediction error.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- Adding trial residual correlation gave no clear gain in a fixed strength prediction test of 52 mouse V1 inhibitory connections with test mice excluded
This prior test assessed mean response similarity with test mice excluded. We test added trial residual information after the mean response measure is given.
- usesVisual similarity gave no error decrease in a split-trial variance test of 75 mouse V1 connections
We use its verified visual matching and selection table, source trial amplitudes for validation, and pinned source receipts. We calculate new fixed-window trial measurements.
- Free cell terms in each direction leave no visual matching contrast in 92 detected mouse V1 connections
This prior study tests visual similarity prediction with measured cell properties and excluded test mice.
- Visual similarity was linked to common voltage strength in 31 selected reciprocal pairs from mouse V1
Prior visual similarity and strength prediction test across excluded mice.