Artificial intelligence for natural product drug discovery.

Developments in computational omics technologies have provided new means to access the hidden diversity of natural products, unearthing new potential for drug discovery. In parallel, artificial intelligence approaches such as machine learning have led to exciting developments in the computational drug design field, facilitating biological activity prediction and de novo drug design for molecular targets of interest. Here, we describe current and future synergies between these developments to effectively identify drug candidates from the plethora of molecules produced by nature. We also discuss how to address key challenges in realizing the potential of these synergies, such as the need for high-quality datasets to train deep learning algorithms and appropriate strategies for algorithm validation.

Comprehensive prediction of secondary metabolite structure and biological activity from microbial genome sequences.

Novel antibiotics are urgently needed to address the looming global crisis of antibiotic resistance. Historically, the primary source of clinically used antibiotics has been microbial secondary metabolism. Microbial genome sequencing has revealed a plethora of uncharacterized natural antibiotics that remain to be discovered. However, the isolation of these molecules is hindered by the challenge of linking sequence information to the chemical structures of the encoded molecules. Here, we present PRISM 4, a comprehensive platform for prediction of the chemical structures of genomically encoded antibiotics, including all classes of bacterial antibiotics currently in clinical use. The accuracy of chemical structure prediction enables the development of machine-learning methods to predict the likely biological activity of encoded molecules. We apply PRISM 4 to chart secondary metabolite biosynthesis in a collection of over 10,000 bacterial genomes from both cultured isolates and metagenomic datasets, revealing thousands of encoded antibiotics. PRISM 4 is freely available as an interactive web application at http://prism.adapsyn.com .

DeepRiPP integrates multiomics data to automate discovery of novel ribosomally synthesized natural products.

Microbial natural products represent a rich resource of evolved chemistry that forms the basis for the majority of pharmacotherapeutics. Ribosomally synthesized and posttranslationally modified peptides (RiPPs) are a particularly interesting class of natural products noted for their unique mode of biosynthesis and biological activities. Analyses of sequenced microbial genomes have revealed an enormous number of biosynthetic loci encoding RiPPs but whose products remain cryptic. In parallel, analyses of bacterial metabolomes typically assign chemical structures to only a minority of detected metabolites. Aligning these 2 disparate sources of data could provide a comprehensive strategy for natural product discovery. Here we present DeepRiPP, an integrated genomic and metabolomic platform that employs machine learning to automate the selective discovery and isolation of novel RiPPs. DeepRiPP includes 3 modules. The first, NLPPrecursor, identifies RiPPs independent of genomic context and neighboring biosynthetic genes. The second module, BARLEY, prioritizes loci that encode novel compounds, while the third, CLAMS, automates the isolation of their corresponding products from complex bacterial extracts. DeepRiPP pinpoints target metabolites using large-scale comparative metabolomics analysis across a database of 10,498 extracts generated from 463 strains. We apply the DeepRiPP platform to expand the landscape of novel RiPPs encoded within sequenced genomes and to discover 3 novel RiPPs, whose structures are exactly as predicted by our platform. By building on advances in machine learning technologies, DeepRiPP integrates genomic and metabolomic data to guide the isolation of novel RiPPs in an automated manner.

Global analysis of prokaryotic tRNA-derived cyclodipeptide biosynthesis.

BACKGROUND: Among naturally occurring small molecules, tRNA-derived cyclodipeptides are a class that have attracted attention for their diverse and desirable biological activities. However, no tools are available to link cyclodipeptide synthases identified within prokaryotic genome sequences to their chemical products. Consequently, it is unclear how many genetically encoded cyclodipeptides represent novel products, and which producing organisms should be targeted for discovery. RESULTS: We developed a pipeline for identification and classification of cyclodipeptide biosynthetic gene clusters and prediction of aminoacyl-tRNA substrates and complete chemical structures. We leveraged this tool to conduct a global analysis of tRNA-derived cyclodipeptide biosynthesis in 93,107 prokaryotic genomes, and compared predicted cyclodipeptides to known cyclodipeptide synthase products and all known chemically characterized cyclodipeptides. By integrating predicted chemical structures and gene cluster architectures, we created a unified map of known and unknown genetically encoded cyclodipeptides. CONCLUSIONS: Our analysis suggests that sizeable regions of the chemical space encoded within sequenced prokaryotic genomes remain unexplored. Our map of the landscape of genetically encoded cyclodipeptides provides candidates for targeted discovery of novel compounds. The integration of our pipeline into a user-friendly web application provides a resource for further discovery of cyclodipeptides in newly sequenced prokaryotic genomes.

Comparative analysis of chemical similarity methods for modular natural products with a hypothetical structure enumeration algorithm.

Natural products represent a prominent source of pharmaceutically and industrially important agents. Calculating the chemical similarity of two molecules is a central task in cheminformatics, with applications at multiple stages of the drug discovery pipeline. Quantifying the similarity of natural products is a particularly important problem, as the biological activities of these molecules have been extensively optimized by natural selection. The large and structurally complex scaffolds of natural products distinguish their physical and chemical properties from those of synthetic compounds. However, no analysis of the performance of existing methods for molecular similarity calculation specific to natural products has been reported to date. Here, we present LEMONS, an algorithm for the enumeration of hypothetical modular natural product structures. We leverage this algorithm to conduct a comparative analysis of molecular similarity methods within the unique chemical space occupied by modular natural products using controlled synthetic data, and comprehensively investigate the impact of diverse biosynthetic parameters on similarity search. We additionally investigate a recently described algorithm for natural product retrobiosynthesis and alignment, and find that when rule-based retrobiosynthesis can be applied, this approach outperforms conventional two-dimensional fingerprints, suggesting it may represent a valuable approach for the targeted exploration of natural product chemical space and microbial genome mining. Our open-source algorithm is an extensible method of enumerating hypothetical natural product structures with diverse potential applications in bioinformatics.

Polyketide and nonribosomal peptide retro-biosynthesis and global gene cluster matching.

Polyketides (PKs) and nonribosomal peptides (NRPs) are profoundly important natural products, forming the foundations of many therapeutic regimes. Decades of research have revealed over 11,000 PK and NRP structures, and genome sequencing is uncovering new PK and NRP gene clusters at an unprecedented rate. However, only ∼10% of PK and NRPs are currently associated with gene clusters, and it is unclear how many of these orphan gene clusters encode previously isolated molecules. Therefore, to efficiently guide the discovery of new molecules, we must first systematically de-orphan emergent gene clusters from genomes. Here we provide to our knowledge the first comprehensive retro-biosynthetic program, generalized retro-biosynthetic assembly prediction engine (GRAPE), for PK and NRP families and introduce a computational pipeline, global alignment for natural products cheminformatics (GARLIC), to uncover how observed biosynthetic gene clusters relate to known molecules, leading to the identification of gene clusters that encode new molecules.

Genomic charting of ribosomally synthesized natural product chemical space facilitates targeted mining.

Microbial natural products are an evolved resource of bioactive small molecules, which form the foundation of many modern therapeutic regimes. Ribosomally synthesized and posttranslationally modified peptides (RiPPs) represent a class of natural products which have attracted extensive interest for their diverse chemical structures and potent biological activities. Genome sequencing has revealed that the vast majority of genetically encoded natural products remain unknown. Many bioinformatic resources have therefore been developed to predict the chemical structures of natural products, particularly nonribosomal peptides and polyketides, from sequence data. However, the diversity and complexity of RiPPs have challenged systematic investigation of RiPP diversity, and consequently the vast majority of genetically encoded RiPPs remain chemical "dark matter." Here, we introduce an algorithm to catalog RiPP biosynthetic gene clusters and chart genetically encoded RiPP chemical space. A global analysis of 65,421 prokaryotic genomes revealed 30,261 RiPP clusters, encoding 2,231 unique products. We further leverage the structure predictions generated by our algorithm to facilitate the genome-guided discovery of a molecule from a rare family of RiPPs. Our results provide the systematic investigation of RiPP genetic and chemical space, revealing the widespread distribution of RiPP biosynthesis throughout the prokaryotic tree of life, and provide a platform for the targeted discovery of RiPPs based on genome sequencing.

Assembly and clustering of natural antibiotics guides target identification.

Antibiotics are essential for numerous medical procedures, including the treatment of bacterial infections, but their widespread use has led to the accumulation of resistance, prompting calls for the discovery of antibacterial agents with new targets. A majority of clinically approved antibacterial scaffolds are derived from microbial natural products, but these valuable molecules are not well annotated or organized, limiting the efficacy of modern informatic analyses. Here, we provide a comprehensive resource defining the targets, chemical origins and families of the natural antibacterial collective through a retrobiosynthetic algorithm. From this we also detail the directed mining of biosynthetic scaffolds and resistance determinants to reveal structures with a high likelihood of having previously unknown modes of action. Implementing this pipeline led to investigations of the telomycin family of natural products from Streptomyces canus, revealing that these bactericidal molecules possess a new antibacterial mode of action dependent on the bacterial phospholipid cardiolipin.

Genomes to natural products PRediction Informatics for Secondary Metabolomes (PRISM).

Microbial natural products are an invaluable source of evolved bioactive small molecules and pharmaceutical agents. Next-generation and metagenomic sequencing indicates untapped genomic potential, yet high rediscovery rates of known metabolites increasingly frustrate conventional natural product screening programs. New methods to connect biosynthetic gene clusters to novel chemical scaffolds are therefore critical to enable the targeted discovery of genetically encoded natural products. Here, we present PRISM, a computational resource for the identification of biosynthetic gene clusters, prediction of genetically encoded nonribosomal peptides and type I and II polyketides, and bio- and cheminformatic dereplication of known natural products. PRISM implements novel algorithms which render it uniquely capable of predicting type II polyketides, deoxygenated sugars, and starter units, making it a comprehensive genome-guided chemical structure prediction engine. A library of 57 tailoring reactions is leveraged for combinatorial scaffold library generation when multiple potential substrates are consistent with biosynthetic logic. We compare the accuracy of PRISM to existing genomic analysis platforms. PRISM is an open-source, user-friendly web application available at http://magarveylab.ca/prism/.

The Cytochrome P450 superfamily complement (CYPome) in the annelid Capitella teleta.

The Cytochrome P450 super family (CYP) is responsible for a wide range of functions in metazoans, having roles in both exogenous and endogenous substrate metabolism. Annelids are known to metabolize polycyclic aromatic hydrocarbons (PAHs) and produce estrogen. CYPs are postulated to be key enzymes in these processes in annelids. In this study, the CYP complement (CYPome) of the annelid Capitella teleta has been robustly identified and annotated with the genome assembly available. Phylogenetic analyses were performed to understand the evolutionary relationships between CYPs in C. teleta and other species. Predictions of which CYPs are potentially involved in both PAH metabolism and steroidogensis were made based on phylogeny. Annotation of 84 full length and 12 partial CYP sequences predicted a total of 96 functional CYPs in C. teleta. A further 13 CYP fragments were found but these may be pseudogenes. The C. teleta CYPome contained 24 novel CYP families and seven novel CYP subfamilies within existing families. A phylogenetic analysis identified that the C. teleta sequences were found in 9 of the 11 metazoan CYP clans. Two CYPs, CYP3071A1 and CYP3072A1, did not cluster with any metazoan CYP clans. We found xenobiotic response elements (XREs) upstream of C. teleta CYPs related to vertebrate CYP1 (CYP3060A1, CYP3061A1) and from families with reported transcriptional upregulation in response to PAH exposure (CYP4, CYP331). C. teleta had a CYP51A1 with ∼65% identity to vertebrate CYP51A1 sequences and has been predicted to have lanosterol 14 α-demethylase activity. CYP376A1, CYP3068A1, CYP3069A1, and CYP3070A1 were the most appropriate candidates for steroidogenesis genes based on their phylogeny and warrant further analyses, though no specific aromatase (estrogen synthesis) candidates were found. Presence of XREs upstream of C. teleta CYPs may indicate a functional aryl hydrocarbon receptor in C. teleta and candidate CYPs for studies of PAH metabolism.