DeepSeMS

DeepSeMS predicts microbial secondary metabolite structures from biosynthetic gene clusters using a large language model built on the Transformer architecture.

Key Features:

• Functional domain-based sequence feature identification: DeepSeMS identifies sequence features using functional domains of biosynthetic enzymes.
• Feature-aligned chemical structure enumeration: The model incorporates feature-aligned chemical structure enumeration for training data augmentation.
• Cryptic biosynthetic gene cluster prediction: DeepSeMS successfully predicted secondary metabolites for 96.60% of cryptic biosynthetic gene clusters.
• Integrated global ocean microbial secondary metabolites repository: The platform includes a repository of over 65,000 novel secondary metabolites from the global ocean microbiome.

Scientific Applications:

• Secondary metabolite structure prediction: DeepSeMS predicts chemical structures of secondary metabolites from various microbial biosynthetic gene clusters with Tanimoto coefficients up to 0.6 compared with ground truth.
• Global ocean microbiome biosynthetic potential discovery: DeepSeMS characterized over 65,000 novel secondary metabolites with previously undocumented structural types, ecological distribution, and biomedical applications especially antibiotics.
• Comparative performance benchmarking: DeepSeMS significantly outperforms antiSMASH (Tanimoto coefficient 0.14) and PRISM (Tanimoto coefficient 0.45) in chemical structure prediction accuracy.

Methodology:

DeepSeMS is built on the Transformer architecture and uses functional domains of biosynthetic enzymes for sequence feature identification, with feature-aligned chemical structure enumeration for training data augmentation.