(A) Median response spectra of all functionally-identified glomeruli. Each spectrum (row) is normalized to its maximal odorant response. Odorants ordered according to nominal structural classification (see Materials and methods), as in Figure 1E. (B) Median response spectra of functionally-identified glomeruli, with odorants ordered by physicochemical descriptor distance from the primary odorant for each glomerulus (odorant that evokes a response at the lowest concentration), using the Dragon1377 descriptor set. (C) Cumulative fraction of all odorant responses encountered with increasing ranked distance from the primary odorant, averaged across all 19 functionally identified glomeruli with co-tuning (i.e. responding to more than one odorant). Colored lines show the cumulative response fraction observed for different odorant rankings, including distance in: the Dragon1377 descriptor set (as shown in (B)), the Dragon1377 descriptor set re-ordered relative to the strongest-activating odorant (Dragon1377 (max)), and a larger 2982-element descriptor set (Alvad. 2982). Black line shows the cumulative response fraction of binarized odorant responses, with odorants ranked by response magnitude, representing the maximum achievable response prediction by odorant ranking. Grey lines show cumulative response fraction following random odorant ranking (100 iterations), representing chance prediction. (D) Same as (C) for all co-tuned glomeruli imaged across the eight OBs (n=694), relative to the primary odorant for each glomerulus, for different odorant rankings, including distance in: the Dragon1377 descriptor set, a previously optimized subset of 32 Dragon descriptors (Dragon32), a subset of descriptors defining functional groups (FG(49)), a subset of previously developed chemical feature binarized fingerprints (i.e. bits; MACCS(106)), and a novel set of chemical feature binarized fingerprints (SMARTS42). Grey line shows mean cumulative response fraction following random odorant ranking. (E) Distribution of performance metrics (see Materials and methods; 1 indicates perfect prediction, 0 indicates chance prediction) across all 694 co-tuned glomeruli for different odorant rankings. Red bar, median; Center bar, interquartile range; envelope, smoothed point density. Asterisks indicate significant difference between odorant rankings (Kruskal-Wallis test comparing all performance metrics: p<2 × 10–16; chi-squared statistic, 859; df = 6; p=0.002 for MACCS(106) vs. Dragon32; p<1 × 10–25 for MACCS(106) vs. all other rankings (post-hoc Dunn tests, Benjamani-Hochberg correction for multiple comparisons); only differences between incrementally higher medians are shown for clarity). (F) Prediction performance metrics for each of the 694 co-tuned glomeruli for the 32-element optimized Dragon descriptor subset (Dragon32; black points) and the 42-element SMARTS fingerprints (SMARTS42; red points), with glomeruli ordered by Dragon32 performance.