PRIMARY ODOR AND MOLECULAR SHAPE 103 Figure 3. Black silhouette from molecular model of isovaleric acid. (Photo courtesy of G. Palmieri and E. Wanke, University of Genoa) The model is photographed in silhouette from three directions mutually at right angles. The silhouette photographs are then sent to Prof. G. Palmieri at the University of Genoa, Italy. He has developed a unique pattern-recog- nition machine, known familiarly as "PAPA." This consists of a modi- fied television camera linked directly to a special computer (9). The machine scans the molecular silhouette by means of a large, reproducible collection of random lines (Fig. 3). The instrument is first "trained" on the silhouette photograph of the isovaleric acid molecule, which .was

104 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS 5 1 o 22 o i 24 26 28 30 32 34 Molecular similarity [Log 2] Figure 4. Correlation of primary odor with molecular shape for 15 fatty acids. Threshold difference for each acid is correlated with the degree to which its molecular shape re- sembles that of isovaleric acid itself (open circle). Both regression lines are shown be- cause both variables are subject to experimental error. (Reproduced from reL 6 by courtesy of the Editor of Nature) selected as the standard compound for the sweaty primary odor. When the machine is subsequently shown photographs of other molecular models, it compares their sizes and shapes by computing the frequencies with which each random line intersects each silhouette. The degree of molecular similarity, assessed by 4,096 random lines, is printed out within five seconds. The final crucial test can now be performed. Do the measurements of primary odor quality correlate with the measurements of molecular shape? Indeed they do (6). The results for 15 fatty acids are given in Fig. 4. The molecular similarity values were those obtained •rom the PAPA machine. The "odor primacy" measurements are the olfactory deficiencies experienced by the specific anosmics, expressed in binary steps. The trend of the correlation is very apparent. The correlation coetficient for this graph is 0.80 (maximum for a straight line would be 1.0). This is a very encouraging result, considering the sources of error and approximation (human variability, problem of chemical purification, simplified molecular modeling). It shows without doubt that there is a strong correlation between molecular shape and odor quality. If it were known that a compound is a fatty acid, it should be possible to make a fair prediction, from its molecular shape, of how much primary sweaty odor character it ought to possess.