Fig. 1 Schematic diagram for sequences and functions of BGC (with penicillin biosynthesis as an example)

Fig. 2 Overall process for BGC mining(This process includes the integration of metagenomic data, prediction of genes and potential BGC, endogenous or heterologous expression, identification of natural products, etc. The case chosen in this figure is Nostocyclopeptide A2, which is extracted from Nostoc sp. ATCC53789 isolated from lichen. It can be used as an inhibitor of 20S proteasome and exhibits anticancer activity[45].)

Fig. 3 Overall flow for BGC analysis and mining[It mainly includes: BGC mining methods (sequence alignment, feature characterization, etc.) and BGC optimization methods (database searching, evolutionary analysis, etc.). Among them, the mining methods of BGC mainly include sequence alignment and feature characterization. Sequence alignment mainly uses BLAST and other methods, while feature characterization employs both traditional methods such as hidden Markov model (HMM) alignment and deep learning based on data model. The optimization methods of BGC mainly include database searching, evolutionary analysis, etc. Database searching includes the searching of BGC sequence database and BGC related small molecule mass spectrometry database, and the main purpose of evolutionary analysis is to analyze the evolution and variation patterns of BGC[54].]

Fig. 4 Analytical methods for establishing correlation between BGC and the production of secondary metabolites[58](a) Retro-biosynthesis: starting with a known compound but no related gene clusters identified, it is possible for predicting enzyme(s) to catalyze the synthesis of such a compound (backbone and tailoring enzymes), and with these predictions putative gene clusters matching the requirements can be found in the genome. The selected case in this figure is penicillin G[59]. (b) Homology searching: starting with a known compound produced by organism 1 and the same or similar compound produced by organism 2 with gene cluster identified, it is possible to use the known gene cluster from organism 2 to search for a similar gene cluster in the genome of organism 1, and thereby identify the gene cluster of interest. (c) Comparative genomics: starting with a group of organisms, some of which produce compounds of interest and some of which do not, it is possible to identify homologous gene clusters in the species that produce them and to screen on the basis of the absence of homologous genes in the species that does not produce them, thereby identifying candidate gene clusters.

Fig. 5 Types of sequence data and corresponding AI analysis methodsDNN— deep neural network; CNN— convolutional neural network; NN— neural network; TL— transfer learning; GCN— graph convolutional network; HMM— hidden markov model

Fig. 6 Status quo and trend of BGC mining using artificial intelligence(Starting from the data, data mining and model construction are carried out with artificial intelligence methods, thus serving the transformation research of synthetic biology, generating more multimodal data and forming a virtuous cycle.)

Fig. 7 Advantages, disadvantages and complementarities of artificial intelligence data mining and culturomics(The list of advantages and disadvantages of the relevant methods is based on the results of comparison with each other and with traditional molecular biological methods as well.)

Fig. 8 BGC's central role in systems biology and synthetic biology(Research on intelligent mining and verification transformation of biosynthetic gene clusters not only connects BGC database with entity database, but also connects artificial intelligence mining and culture experiment verification. Research on intelligent discovery and transformation verification for biosynthetic gene clusters can closely link systems biology and synthetic biology, and realize seamless transformation from data to model and from verification to application.)