DeepAden: an explainable machine learning method for predicting the substrate specificity of nonribosomal peptide synthetases.

Microbial nonribosomal peptides (NRPs) exhibit remarkable structural diversity and are important sources of lead compounds for clinical drug development. The biosynthesis of NRPs relies on nonribosomal peptide synthetases (NRPSs), in which adenylation (A) domains define the core structure by selectively recognizing and activating amino acid substrates. Accurately predicting the substrate specificities of A-domains is thus essential for understanding the core structural and biosynthetic logic of NRPs. Here, we present DeepAden, a two-stage deep learning model. In the first stage, a graph attention network (GAT)-based model localizes 27-residue binding pockets within 6 Å of bound substrates and converts these into pocket representations. In the second stage, pocket representations are encoded alongside substrate information using pretrained language models and aligned using contrastive learning. We also applied a SHapley Additive exPlanations (SHAP)-guided data augmentation strategy to mitigate class imbalance, particularly focusing on nonproteinogenic substrates. DeepAden achieved competitive performance compared with state-of-the-art tools on a benchmark dataset and facilitated the annotation of two Streptomyces NRPS gene clusters by providing supportive A-domain substrate-specificity predictions. DeepAden provides a practical approach for pocket localization and substrate prediction and may facilitate the discovery and characterization of new NRP natural products. The DeepAden web server is available at https://deepnp.site/.