With relay and partner controls, input orientation profiles gave no clear output prediction gain in a ridge test of 1,314 neurons from one MICrONS mouse

We tested whether the orientation preferences of a neuron's input partners help to predict those of its output partners. We used MICrONS minnie65 v1507 and visual tuning estimates from one mouse. We repeated the complete scan of 303,164,967 connection records and recovered 143,765 links among 8,284 matched neurons. The main test used 1,314 pyramidal relay neurons with global orientation selectivity of at least 0.25 and at least five input and five output partners. We removed reciprocal links to keep the two partner sets separate. Nested ridge models controlled relay tuning, cell labels, soma position, model field centers, partner geometry and partner class fractions. Adding the input mean orientation vector cut output profile test error by 0.004%. The conditional 95% spatial block resample range was -0.205% to 0.206%. Four other spatial splits gave gains from -0.272% to 0.002%. The raw path alignment was 0.09598; its cross-fitted residual score was 0.00262, with a range that included zero. A planned lower-selectivity test used 3,111 relays and gave a 0.328% error decrease. Four added split checks at that cutoff gave gains of 0.042% to 0.316%. A wider penalty grid left the main result near zero. These results concern known partner sets, estimated tuning and one sampled graph. They do not test causal signal transfer. The package preserves the selected graph, source provenance, code, figures, all 26 analyses and instructions. A full repeat matched every numerical summary and table.

connectomics · MICrONS · visual cortex · orientation tuning · two-step paths · ridge regression · spatial validation · structure and function