Mixture Analyses of Air-sampled Pollen Extracts Can Accurately Differentiate Pollen Taxa

Abstract

The daily pollen forecast provides crucial information for allergic patients to avoid exposure to specific pollen. Pollen counts are typically measured with air samplers and analyzed with microscopy by trained experts. In contrast, this study evaluated the effectiveness of identifying the component pollens using the metabolites extracted from an air-sampled pollen mixture. Ambient air-sampled pollen from Munich in 2016 and 2017 was visually identified from reference pollens and extracts were prepared. The extracts were lyophilized, rehydrated in optimal NMR buffers, and filtered to remove large proteins. NMR spectra were analyzed for pollen associated metabolites. Regression and decision-tree based algorithms using the concentration of metabolites, calculated from the NMR spectra outperformed algorithms using the NMR spectra themselves as input data for pollen identification. Categorical prediction algorithms trained for low, medium, high, and very high pollen count groups had accuracies of 74% for the tree, 82% for the grass, and 93% for the weed pollen count. Deep learning models using convolutional neural networks performed better than regression models using NMR spectral input, and were the overall best method in terms of relative error and classification accuracy (86% for tree, 89% for grass, and 93% for weed pollen count). This study demonstrates that NMR spectra of air-sampled pollen extracts can be used in an automated fashion to provide taxa and type-specific measures of the daily pollen count.

Keywords: NMR; aerobiology; exposure; metabolomics; mixtures; pollen.

Figure 1-

Pollen Sampled in Munich 2016. Panels A and B show the pollen count / m³ sampled by day of the year (2016) subdivided into A) Tree/Grass/Weed and B) by Taxa. Panel C shows the total pollen for the year sorted by amount.

Figure 2-

Example NMR data. Figure 2 shows an example NMR spectrum of Day 164 in black overlayed with identified metabolites and the metabolite-simulated spectrum in red. DSS is the concentration and chemical shift standard, see Methods.

Figure 3-

Pairwise Correlations. The correlation of the input (calendar day or metabolite) versus the pollen count for A) tree pollen, B) grass pollen, and C) weed pollen.

Figure 4-

Dimensionality reduction of the NMR data. Panels A-C show a t-SNE analysis of the metabolite data colored for high (red), medium (green), and low (blue) pollen count days for tree (A), grass (B), and weed (C). Panels D-E show a PCA plot of the first two principal components of the NMR spectral input colored as above for tree (D), grass (E), and weed (F) pollen counts. The PCA and t-SNE x and y axes are unitless and scaled for maximum dispersion of points.

Figure 5-

Predicted pollen taxa compared to known pollen count. Each graph shows the actual pollen count per m³ for the calendar day with a black circle, and the predicted value from the best model with a red x. The correlation coefficient of the predicted and actual value is shown inset for cross-validated performance. DTree is the sum of all deciduous trees.

Figure 6 -

Correlation of Actual and Predicted Pollen Counts. A, without sample weights for Trees B, with sample weights for Trees. C, Grass; D, Weed. The correlation coefficient of actual versus predicted is inset for leave-one-out cross validation. Dotted vertical green lines indicate boundaries between pollen count groups: L, M, H, VH. Dotted & Dashed diagonal green line indicates perfect agreement of the actual and predicted values.