Auxiliary Material - Spectra
Biological Serum
Defined Mixture Serum
Computer Generated Serum
Biological CSF
Defined Mixture CSF
Computer Generated CSF
Defined Mixture Serum and CSF
Auxiliary Material - Performance
In the paper "Accurate, fully-automated NMR spectral profiling for metabolomics" we report the identification and quantification accuracy of Bayesil with serum and CSF samples.
We have also tested the performance of Bayesil with increasing library size and with samples with increasing number of compounds. Overall, the results show that Bayesil performs quite well regardless of the number of compounds and coverage of the library. We did find that as the mixture size increases beyond the database size, the performance of Bayesil does decrease slightly.
Increasing Library Size
We have conducted this experiment by creating a mixture of 20 metabolites and increasing the database size from 20, 60 and 120 (where all 20 metabolites are in the spectral library). We also repeated the same experiment where only 15/20 (75%) metabolites in the spectrum are in the spectral libraries. The results are as follows:
Mixture Size | Library Size | Overlap | Quantification Accuracy | Identification Accuracy |
---|---|---|---|---|
20 | 20 | 20 | 99% | 100% |
20 | 60 | 20 | 99% | 96% |
20 | 120 | 20 | 99% | 96% |
20 | 20 | 15 | 99% | 95% |
20 | 60 | 15 | 94% | 92% |
20 | 120 | 15 | 88% | 92% |
Increasing Mixture Complexity
We have conducted this experiment by creating mixtures varying in size from 20, 60 and 120 compounds and using a spectral database of 60 compounds (where all metabolites in the 20 and 60 component mixtures are in the spectral library). We also repeated the same experiment with only 75% coverage. The results are as follows:
Mixture Size | Library Size | Overlap | Quantification Accuracy | Identification Accuracy |
---|---|---|---|---|
20 | 60 | 20 | 99% | 96% |
60 | 60 | 60 | 92% | 96% |
120 | 60 | 60 | 66% | 91% |
20 | 60 | 15 | 94% | 92% |
60 | 60 | 45 | 70% | 95% |
120 | 60 | 45 | 56% | 83% |