About DeepSeMS

DeepSeMS is a web server for prediction of microbial secondary metabolite (SM) structures from biosynthetic gene cluster (BGC).

The goal of this tool is to characterize the chemical structures of a natural molecule produced by microbe from BGC sequences that encoding SM. To realize this goal, we trained a deep learning language model for automatically generating chemical sequences of a SM from input BGC sequences, on a refined dataset with sufficient quantity and superior quality.

The model had been evaluated on external validation datasets of 'known BGCs' with experimentally verified SMs for accuracy and 'cryptic BGCs' without chemical structure of SMs for generalization ability. Evaluation results show that DeepSeMS predicted more precise structures of SMs than rule-based tools, and achieved significant improvement on cryptic BGCs for the abundance, complexity and diversity of generated structures.

For more details about DeepSeMS, please see our paper published.