FunBGC: An Intelligent Framework for Fungal Biosynthetic Gene Cluster Identification

Biosynthetic gene clusters (BGCs) in bacterial and fungal genomes encode bioactive secondary metabolites, which play a crucial role in many therapeutic applications in the pharmaceutical industry. The advancements in genome sequencing technologies, as well as the development of intelligent modeling, promote the identification of BGCs. However, a significant number of novel BGCs remain undiscovered. Especially due to the complex diversity of fungal genomes, previous research on fungal BGCs has been limited. Here, we present FunBGC, an intelligent framework to identify BGCs in fungi. The annotated BGCs are represented as Pfam domain sequences as training data. Because of the limited amount of annotated BGCs in fungus, we adopt the BGC data in the bacterial genome to pre-train the model and then fine-tune the model using fungal BGC data. On the publicly accessible real fungal BGC benchmark dataset, we evaluate the performance of our method at both the Pfam domain level and the BGC coverage level, achieving a precision of 0.9199 at the Pfam domain level and a precision of 0.7941 at the BGC level. In the identification of fungal BGCs, FunBGC achieves superior performance compared to the baseline methods DeepBGC and BiGCARP. Furthermore, FunBGC is capable of discovering novel fungal BGCs in unannotated metagenomic data. Overall, FunBGC provides a robust framework for BGC identification, which is significant for the discovery of novel BGCs, exploration of fungal biosynthetic potential, and drug synthesis.