Average linkage of odor tuning did not reach 90% precision for sister pairs in two of three mice

We tested whether odor responses can recover traced common glomerular groups in three anesthetized mice. The source data had 412 olfactory bulb cells, 107 anatomical groups, and 45,447 cell pairs. Pearson correlation of 0 to 2 second mean calcium responses ranked sister pairs with ROC AUC 0.925, 0.864, and 0.876. A fixed average-linkage rule selected a cutoff in two mice and tested it in the third. It required at least 90% pair precision and ten joined pairs in each training mouse. No cutoff qualified in any fold, so the primary rule made no joins. A check added after this result evaluated every merge height within each recording. Maximum precision of any partition with at least one joined pair was 66.7% in Y391 and 88.9% in C432a. These limits did not come from the cutoff grid or pair-count minimum. A secondary rule selected a cutoff to maximize training F1. Test precision was 39.9%, 26.6%, and 78.7%. Corresponding recall was 67.0%, 50.0%, and 26.8%. Only two of 67 nonsingleton anatomical groups were recovered exactly. The strongest Y391 nonsister pair had correlation 0.982 and high similarity in separate repeat halves. Source skeletons confirmed distinct somata and parent labels. These results limit the tested response processing and grouping rules. They do not show that activity cannot reveal anatomy. The package preserves independently extracted inputs, source hashes, code, local plans, all partitions, figures, and repeat instructions. Pair dependence, three mice, different cell classes, fluorescence effects, and incomplete sampling limit general claims.

olfactory bulb · structure and function · sister cells · functional clustering · anatomical validation · calcium imaging · reproducible benchmark