Artificial Intelligence For Natural Product Drug Discovery

AI technologies have been developed to accelerate the discovery of natural products by predicting biosynthetic genes and metabolite structures from sequence or spectral data. Rule-based methods such as those used in antiSMASH and PRISM are successful at detecting known BGC classes, but are less proficient at identifying novel types of BGC or unclustered pathways. Machine learning algorithms, such as hidden Markov model-based ClusterFinder, deep learning approaches DeepBGC, GECCO and SanntiS, and genome mining algorithms for RiPPs, offer significant advantages over rule-based methods. These methods have already demonstrated utility in identifying novel classes of natural product biosynthetic pathways, such as the decRiPPter algorithm which identified pristinin, a novel class of lanthipeptides. Metabolomics allows direct detection of biosynthesized components, even if their precise structures are unknown.