Evaluating Odor Selectivity in α-Synucleinopathy: ML-Based Odor Profiling

• Christian Mattjie PUCRS
• Rafaela Ravazio PUCRS
• Eleanor Mitchell Queen Mary University of London
• Jonathan Bestwick Queen Mary University of London
• Lucas Silveira Kupssinskü PUCRS
• Alastair Noyce Queen Mary University of London
• Artur Schumacher Schuh HCPA
• Rodrigo C. Barros PUCRS

Resumo

Cerebrospinal fluid (CSF) α-synuclein seed-amplification assay (SAA) is the most sensitive in vivo marker of misfolded α-synuclein. Still, its reliance on lumbar puncture—a costly, time-consuming, and invasive procedure—restricts broad clinical or population-based use. Low-cost, non-invasive smell tests such as the 40–item University of Pennsylvania Smell Identification Test (UPSIT) could serve as an efficient triage tool if they could reliably predict CSF-SAA positivity. This work aims to determine if item-level UPSIT responses reveal a selective olfactory pattern of SAA positivity after strict control for overall UPSIT score. Such a pattern could be used to enhance the diagnostic value of smell tests further. UPSIT and SAA data from the Parkinson’s Progression Markers Initiative were used to build two balanced SAA⁺ – SAA^- subsets. Six classifiers were tuned and evaluated using leave-one-out cross-validation. Permutation SHAP values quantified the importance of each odor for the best models. Given the marginal gain over random chance that the models achieved, our findings parallel recent Parkinson’s disease studies, which suggest that SAA causes a general loss of smell without affecting specific smells more than others, or that such a difference is minor. Although odor selectivity was negligible, UPSIT and other olfactory tests remain valuable and scalable screens for potential CSF-SAA positivity, as overall olfactory capacity itself carries strong predictive power.