- Current natural product (NP) research is limited by its reliance on bioassay-guided fractionation to identify bioactive compounds in mixtures. Computational approaches may improve NP research by correlating mass spectra (MS) and nuclear magnetic resonance spectra (NMR) of bioactive mixtures to their bioactivity patterns. In this study, I use artificial neural networks (ANNs) to correlate both MS and NMR data of 40 fractions of hops (Humulus lupulus) extract to inhibition of iNOS-mediated formation of nitric oxide (provided by the Stevens Lab at Oregon State University). Xanthohumol and its derivatives, constituents of hops extract, are known to exhibit this anti-inflammatory activity. An MS-based model, NMR-based model, a model that concatenates MS and NMR as a single input, and model that treats them as separate inputs were investigated. The MS-based model predicted bioactivity with lowest error (MSE = 0.685) and identified xanthohumol as the top anti-inflammatory compound. The NMR-based model, concatenated model, and multichannel model predicted bioactivity with higher error: MSE = 9.48, 8.05, and 7.58, respectively, but they identified several known bioactive molecules and associated proton shifts as top predictors. In conclusion, ANNs have been shown to usefully predict bioactivity from MS/NMR data.
- Keywords: Machine learning, artificial neural networks, mass spectra, nuclear magnetic resonance spectra, bioactivity, natural products, Humulus lupulus, xanthohumol