Fig. 1. Results of principal component analysis for 80,000 spores per sample, 100 samples for each species.

Fig. 2. Decision tree analysis using data from 80,000 spores per sample, 100 samples for each species.

Fig. 3. Distribution of features across 40 models. (a) B. subtilis vs. B. thuringiensis, (b) B. subtilis vs. B. cereus, and (c) B. cereus vs. B. thuringiensis.

Fig. 4. Plot of the dominant classifier feature 18097. The intensity of this biomarker feature was extracted from each of the 100 raw data files for the 5k concentration of B. subtilis (+) and B. thuringiensis (○). The data for each species shows a different distribution at this point. The classification algorithm finds data points such as this to aid in decision-making. While one feature alone cannot completely discriminate the two species, the unique combination of features within a model does.

Fig. 5. Representative differential mobility spectra of 80,000 spores undergoing pyrolysis at 650 °C for 99.99 s. Positive ion spectrum left, negative ion spectrum right. X-axis represents Vc (V), Y-axis represents scan number. The nine features from three-way model (a) are circled in black. (a) B. subtilis, (b) B. cereus, and (c) B. thuringiensis.

Comparisons of validation results based on two-way modeling across all concentrations (80k spores, 10k, and 5k)

One hundred one B. cereus, 99 B. subtilis, and 100 B. thuringiensis files were used. Data shown for each binary comparison include number of biomarker features (B), decision boundary (DB), number of nodes (N), sensitivity (Sn), specificity (Sp), and percent accuracy (A). Sensitivity and specificity are calculated with respect to the first species named in each comparison.

Comparison of validation results from modeling across the three spore concentrations B. subtilis vs. B. cereus and B. thuringiensis together

One hundred files of B. subtilis were modeled against 200 files of the other species (100 files each of B. cereus and B. thuringiensis). Abbreviations as in Table 1. Sensitivity and specificity are calculated with respect to B. subtilis.

Results of modeling B. cereus vs. B. subtilis vs. B. thuringiensis in a single three-way model

Twenty five files of each species at the 80k concentration were used for validation. Reading down the columns, one can determine how those 25 files were classified. Two models are shown. Model (a) using a decision boundary setting of 0.9, contains 9 features and 22 clusters, with an overall accuracy of 77.3%. Model (b) using a decision boundary setting of 0.8, contains 12 features and 4 clusters, with an overall accuracy of 73.3%.